Chapter VI

Receivers & Transmitters

Transceiver Architectures — Heterodyne, Zero-IF, Image-Reject (Hartley/Weaver), Low-IF; Direct-Conversion & Two-Step Transmitters; I/Q mismatch, DC offset, LO pulling

Outline — click to expand
6.1 General Considerations
6.2 Receiver Architectures
6.3 Transmitter Architectures
6.4 Case Studies & Summary
Practice Problems

6.1 General Considerations

What we optimize in a transceiver

A radio transceiver has two signal chains. The receiver (RX) must pull a tiny desired signal out of a sea of strong interferers; the transmitter (TX) must deliver clean, efficient power without polluting adjacent channels.

Receiver metrics

Sensitivity · Linearity · Dynamic range · Gain · Power dissipation · Complexity. The receiver must detect a weak desired channel while large interferers sit nearby.

Transmitter metrics

Output power · Spurious emission · Linearity · Efficiency · Power dissipation · Complexity. The transmitter must not leak energy into adjacent channels.

RX front-end interferers desired BPF LNA → mixer… TX front-end …mixer → PA BPF transmitted adjacent ch.
RX: dig the desired channel out from interferers. TX: deliver power into the channel without spilling into adjacent channels. (Slide 3)

Why every stage before channel-select must be linear

The channel-select filter is the first place where interferers are actually removed. Everything before it — antenna BPF, LNA, mixer — sees the full interferer power. If those stages are nonlinear, two strong interferers at $\omega_1,\omega_2$ generate third-order intermodulation products at $2\omega_1-\omega_2$ and $2\omega_2-\omega_1$ that can fall directly on the desired channel and cannot be filtered out afterward.

Design rule
All stages in the RX chain that precede channel-select filtering must be sufficiently linear — their IP3 sets the receiver's blocker tolerance. This is why LNA and mixer IP3 are first-order specs.

LNA desensitization by the PA

In a full-duplex radio (TX and RX on simultaneously through a duplexer), the PA's large output (tens of volts of swing) leaks into the RX input and can desensitize the LNA (compress its gain). GSM avoids this by time-offsetting the RX and TX slots (TDD-like) so the PA is off while the receiver listens. (Slide 4)

GSM avoids this issue by offsetting the RX and TX time slots. antenna Duplexer recv and send filter LNA PA 20 V 少量跑過來——因為 PA 功率實在太大a small amount still gets through — because the PA power is simply enormousわずかに漏れ込む——PA の電力が桁違いに大きいため
Slide 4:全雙工(recv and send 同時)經 duplexer 共用天線。PA 輸出擺幅 ~20 V,即使 duplexer 有 filter 隔離,仍有少量洩漏進 RX 端(灰色路徑)→ Slide 4: full duplex (receiving and transmitting at once) sharing one antenna through a duplexer. The PA output swings ~20 V, and even with the duplexer's filter isolation a small amount still leaks into the RX side (the grey path) → スライド 4:全二重(受信と送信を同時に行う)で、デュプレクサを介して 1 本のアンテナを共用する。PA 出力の振幅は約 20 V あり、デュプレクサのフィルタで分離してもわずかな量が RX 側へ漏れ込む(灰色の経路)→ LNA 增益壓縮(desensitization)LNA gain compression (desensitization)LNA の利得圧縮(desensitization);接收的微弱訊號(上方小波)就被蓋掉。; the weak received signal (the small wave at the top) is swamped.;受信した微弱信号(上部の小さな波)はかき消されてしまう。GSM 用錯開 RX/TX time slots 避開此問題GSM avoids the problem by staggering the RX/TX time slotsGSM は RX/TX のタイムスロットをずらすことでこの問題を回避する——收的時候 PA 根本沒開。— when it is receiving, the PA is simply off.——受信中は PA がそもそもオフになっている。

6.2 Receiver Architectures

Heterodyne (single-conversion) architecture

You cannot build a sharp, tunable channel-select filter directly at RF (a 200 kHz channel at 2 GHz needs $Q\!\sim\!10^4$). The heterodyne solution: translate the RF channel down to a fixed, lower intermediate frequency (IF) where high-Q filtering is feasible.

ω₁ LNA A₀ cos ω₀t LPF ω₂ (IF) 高 ω → 低 ωhigh ω → low ω高い ω → 低い ω translate down to IF
Single downconversion: LNA → mixer (×LO) → LPF moves the RF channel to a fixed IF. (Slide 5)
Heterodyne architecture and the image problem (slide)
Slides 原圖:上半 — (a) 把 RF 通道用 $A_0\cos\omega_0 t$「translate」到較低頻率(RF 端濾不掉 interferer),(b) 實際鏈路 LNA → mixer → LPF。下半 — Original slide figures: top — (a) the RF channel is “translated” down in frequency by $A_0\cos\omega_0 t$ (the interferer cannot be filtered at RF); (b) the actual chain, LNA → mixer → LPF. Bottom — スライド原図:上段 — (a) RF チャネルを $A_0\cos\omega_0 t$ で低い周波数へ「translate」する(RF 段では干渉波を濾波できない)、(b) 実際の信号経路 LNA → mixer → LPF。下段 — Problem of Image:desired band 與 image 對稱分佈在 $\omega_{LO}$ 兩側、各距 $\omega_{IF}$;混頻後兩者都落在 $\omega_{IF}$,image 疊在訊號上(右端頻譜黑色尖峰),詳見下一節。: the desired band and the image sit symmetrically on either side of $\omega_{LO}$, each $\omega_{IF}$ away; after mixing both land on $\omega_{IF}$ and the image lands on top of the signal (the black spike in the spectrum at the right). See the next section.:希望波帯域とイメージが $\omega_{LO}$ の両側に対称に位置し、それぞれ $\omega_{IF}$ だけ離れている;ミキシング後は両者とも $\omega_{IF}$ に落ち、イメージが信号に重なる(右端スペクトルの黒いピーク)。詳しくは次節を参照。

The image problem

A mixer multiplies by $\cos\omega_{LO}t$, which has spectral content at both $+\omega_{LO}$ and $-\omega_{LO}$. So two input frequencies map to the same IF: the desired channel at $\omega_{LO}+\omega_{IF}$ and the image at $\omega_{LO}-\omega_{IF}$ (or vice-versa). They sit symmetrically about the LO, separated by $2\omega_{IF}$.

Desired band ω₁ ω_LO Image ω_im ω_IF ω_IF image & desired are 2ω_IF apart, mirrored about ω_LO both fold onto the same IF → ω_IF (both!)
The image and the desired signal are mirror images about the LO. After mixing, both land on the IF — the image corrupts the desired channel unless removed beforehand. (Slide 5)
Image frequency $$\omega_{im} = \omega_{LO} + \omega_{IF} \quad\text{(if signal at }\omega_{LO}-\omega_{IF}\text{)},\qquad |\omega_{sig}-\omega_{im}| = 2\omega_{IF}$$
怎麼來的:$\cos\omega_{in}t\cdot\cos\omega_{LO}t = \tfrac{1}{2}\cos(\omega_{in}-\omega_{LO})t + \tfrac{1}{2}\cos(\omega_{in}+\omega_{LO})t$,LPF 後留下差頻 → 輸出在 $|\omega_{in}-\omega_{LO}|$。要落在 IF 就是解 $|\omega_{in}-\omega_{LO}| = \omega_{IF}$——Where it comes from: $\cos\omega_{in}t\cdot\cos\omega_{LO}t = \tfrac{1}{2}\cos(\omega_{in}-\omega_{LO})t + \tfrac{1}{2}\cos(\omega_{in}+\omega_{LO})t$; the LPF keeps the difference term, so the output is at $|\omega_{in}-\omega_{LO}|$. Landing on the IF means solving $|\omega_{in}-\omega_{LO}| = \omega_{IF}$ —導出:$\cos\omega_{in}t\cdot\cos\omega_{LO}t = \tfrac{1}{2}\cos(\omega_{in}-\omega_{LO})t + \tfrac{1}{2}\cos(\omega_{in}+\omega_{LO})t$ であり、LPF が差の項を残すので出力は $|\omega_{in}-\omega_{LO}|$ に現れる。IF に落ちる条件は $|\omega_{in}-\omega_{LO}| = \omega_{IF}$ を解くことであり——二次絕對值方程有兩個根an absolute-value equation has two roots絶対値を含む方程式には根が 2 つある:$\omega_{LO}-\omega_{IF}$(訊號)和 $\omega_{LO}+\omega_{IF}$(image),關於 $\omega_{LO}$ 鏡像對稱,因此相距 $2\omega_{IF}$: $\omega_{LO}-\omega_{IF}$ (the signal) and $\omega_{LO}+\omega_{IF}$ (the image), mirror-symmetric about $\omega_{LO}$ and therefore $2\omega_{IF}$ apart:$\omega_{LO}-\omega_{IF}$(信号)と $\omega_{LO}+\omega_{IF}$(イメージ)であり、$\omega_{LO}$ を中心に鏡像対称、したがって $2\omega_{IF}$ 離れている

Image-reject filter & the IF trade-off

A more complete heterodyne RX inserts an image-reject filter between the LNA and the mixer to suppress the image band, then a channel-select filter at IF to pick out the wanted channel.

LNA Image Reject Filter cos ω_LO t Channel Select Filter → IF amp image filter at RF, channel filter at IF
LNA → image-reject filter (RF) → mixer → channel-select filter (IF). (Slide 6)
Image reject filter passband (slide)
Slides 原圖:image-reject filter(虛線)通帶罩住 desired band(黃),阻帶壓住 $2\omega_{IF}$ 外的 image(黑)——Original slide figure: the passband of the image-reject filter (dashed) covers the desired band (yellow) while its stopband suppresses the image $2\omega_{IF}$ away (black) —スライド原図:イメージ除去フィルタ(破線)の通過帯域が希望波帯域(黄)を覆い、阻止帯域が $2\omega_{IF}$ 離れたイメージ(黒)を抑圧する——中頻頻率大,image 離得遠,濾波器好設計a high IF puts the image far away, which makes the filter easy to design中間周波数が高いとイメージが遠くなり、フィルタの設計が容易になる(slides 註記);(as annotated on the slides); (スライドの注記); 這正是下方 trade-off 裡「High IF → easy image rejection」那一側this is exactly the “High IF → easy image rejection” side of the trade-off belowこれはまさに下のトレードオフにおける「High IF → easy image rejection」の側である.
The fundamental trade-off
The image sits $2\omega_{IF}$ away from the signal.
  • High IF → image is far away → easy image rejection, but the IF channel-select filter must now be sharp at a high frequency → hard channel selection.
  • Low IF → image is close to the signal → hard image rejection, but channel selection at low IF is easy.
You cannot win both with a single conversion — hence dual conversion, or image-reject mixers.
High-IF vs Low-IF trade-off (slide)
Slides:同一張頻譜看 IF 取捨。(a) Slides: the same spectrum viewed as the IF trade-off. (a) スライド:同じスペクトルを IF のトレードオフとして見る。(a) 高 IFHigh IF高い IF:image 離訊號 $2\omega_{IF}$ 很遠,image-reject filter 輕鬆壓掉;但鄰近通道 interferer(鄰近通道)混到 IF 後緊貼著訊號,channel-select filter 難以濾乾淨(鄰近通道仍在)。(b) : the image is a long way ($2\omega_{IF}$) from the signal and the image-reject filter suppresses it easily; but after mixing, an adjacent-channel interferer sits right next to the signal at IF and the channel-select filter struggles to clean it up (the adjacent channel survives). (b) :イメージは信号から $2\omega_{IF}$ と大きく離れており、イメージ除去フィルタで容易に抑圧できる;しかし隣接チャネルの干渉波はミキシング後 IF で信号のすぐ隣に位置し、チャネル選択フィルタでは除去しきれない(隣接チャネルが残る)。(b) 低 IFLow IF低い IF(用低中頻的狀況):channel selection 容易,但 image 緊貼訊號、落在 image-reject filter 通帶內 → 混頻後 image 也跟著進來(image 仍濾不掉)。(the low-IF case): channel selection is easy, but the image is close to the signal and falls inside the image-reject filter's passband → after mixing the image comes through too (the image still cannot be filtered).(低 IF の場合):チャネル選択は容易だが、イメージが信号に近くイメージ除去フィルタの通過帯域内に入る → ミキシング後はイメージも一緒に入ってくる(イメージは依然として除去できない)。

Dual downconversion(改善上面單次 conversion 問題)Dual downconversion (fixing the single-conversion problem above)ダブルダウンコンバージョン(上記のシングルコンバージョンの問題を改善する)

Splitting the translation into two steps resolves the trade-off: a first high IF gives good image rejection, a second low IF gives good channel selection.

Dual downconversion chain A-H (slide)
Slides 原圖:dual downconversion 完整鏈 — band-select BPF₁ → LNA → image-reject BPF₂ → 1st mix(ω_LO1,高中頻率調 image)→ channel-select BPF₃ → 2nd mix(ω_LO2,低中頻率調鄰近通道)→ channel-select BPF₄ → IF amp;節點 A–H 對應下圖各張頻譜。(Slide 7)Original slide figure: the complete dual-downconversion chain — band-select BPF₁ → LNA → image-reject BPF₂ → 1st mix (ω_LO1, a high IF to deal with the image) → channel-select BPF₃ → 2nd mix (ω_LO2, a low IF to deal with the adjacent channel) → channel-select BPF₄ → IF amp; nodes A–H correspond to the spectra in the figure below. (Slide 7)スライド原図:ダブルダウンコンバージョンの全体構成 — band-select BPF₁ → LNA → image-reject BPF₂ → 1st mix(ω_LO1、イメージに対処するための高い IF)→ channel-select BPF₃ → 2nd mix(ω_LO2、隣接チャネルに対処するための低い IF)→ channel-select BPF₄ → IF amp;ノード A–H は下図の各スペクトルに対応する。(Slide 7)
Dual downconversion spectrum progression A-H (slide)
Slides 原圖(gen1/2 主要技術):頻譜走過整條鏈。A→B band-select;C→D 第一次混頻(高 IF₁,高中頻率調 image:image 被 image-reject filter 壓掉);E→F 第一次 channel-select;G→H 第二次混頻到低 IF₂(低中頻率調鄰近通道:鄰近通道由低 IF 的銳利濾波器處理)。對應上方方塊圖各節點。Original slide figure (the main technique of gen 1/2): the spectrum walked along the whole chain. A→B band select; C→D the first mix (high IF₁, a high IF to deal with the image: the image is suppressed by the image-reject filter); E→F the first channel select; G→H the second mix down to a low IF₂ (a low IF to deal with the adjacent channel: the adjacent channel is handled by a sharp filter at the low IF). These correspond to the nodes of the block diagram above.スライド原図(第 1/2 世代の主要技術):スペクトルが経路全体をたどる様子。A→B バンド選択;C→D 1 回目のミキシング(高い IF₁、イメージに対処するための高い IF:イメージはイメージ除去フィルタで抑圧される);E→F 1 回目のチャネル選択;G→H 低い IF₂ への 2 回目のミキシング(隣接チャネルに対処するための低い IF:隣接チャネルは低 IF の急峻なフィルタで処理する)。上の系統図の各ノードに対応する。
📝 HW11 — Secondary image & the Zero-IF divider questions

(a) Dual downconversion (slide 7): "How about the secondary image?"

(b) Modern heterodyne / Zero-IF with ÷2 divider (slide 8): a single LO at $\omega_{LO}$ drives a ÷2 to make quadrature ($\cos$, $\sin$) at $\omega_{LO}/2$ feeding the I/Q mixers. For a desired band $[\,f_a, f_b\,]$, answer: (1) LO frequency range? (2) IF range? (3) image frequency range? (4) repeat for divide-by-4.

LNA cos ωLOt I Q ÷ 2 直接降到基頻straight down to baseband直接ベースバンドへ ??
Slide 8 的 modern heterodyne RX:LNA 後先與 $\omega_{LO}$ 混頻,再由 The modern heterodyne RX of slide 8: after the LNA the signal is mixed with $\omega_{LO}$, and a スライド 8 の modern heterodyne RX:LNA の後にまず $\omega_{LO}$ と混合し、次に ÷2 產生 $\omega_{LO}/2$ 的正交 LO 給 I/Q 混頻器——$\omega_{RF} = \omega_{LO} + \omega_{LO}/2$,一次降到基頻(zero IF);黃色 ÷2 與「??」就是本題要回答的部分。 then produces a quadrature LO at $\omega_{LO}/2$ for the I/Q mixers — $\omega_{RF} = \omega_{LO} + \omega_{LO}/2$, going down to baseband in one step (zero IF); the yellow ÷2 and the “??” are what this question asks you to fill in. が $\omega_{LO}/2$ の直交 LO を生成して I/Q ミキサへ供給する——$\omega_{RF} = \omega_{LO} + \omega_{LO}/2$ で、一度でベースバンドまで降ろす(zero IF);黄色の ÷2 と「??」が本問で答えるべき部分である。
點此展開解答 · Show answerClick to expand the answer · Show answerクリックして解答を表示 · Show answer

(a) Secondary image. In a two-LO receiver, the second mixer also has its own image. After the first downconversion, signals at $\omega_{IF1}$ and at $2\omega_{LO2}-\omega_{IF1}$ (mirror about $\omega_{LO2}$) both fold onto IF₂. The first channel-select filter (BPF₃) must reject this secondary-image band before the second mixer. The secondary image is the band located $2\omega_{IF2}$ away from the desired signal at the IF₁ plane; BPF₃ provides its rejection.

  1. LO frequency range — for zero-IF the divided LO equals the RF: $f_{LO,div}=f_{RF}$, so the VCO runs at $2f_{RF}$, i.e. $\omega_{LO}\in[2f_a,\,2f_b]$.
  2. IF range — zero-IF ⇒ IF $= 0$ (baseband, channel bandwidth only).
  3. Image frequency range — with $\omega_{IF}=0$ the image is the signal's own negative-frequency twin; I/Q processing separates them, so there is no distinct external image band.
  4. Divide-by-4 case — VCO must run at $4f_{RF}$; quadrature still generated by the divider; same zero-IF baseband result. Running the VCO far from the RF (×4) is precisely what suppresses LO pulling/self-mixing.

Goals of the modern architecture (slide 8): avoid the secondary image, use a single LO, avoid off-chip filters, perform detection in the digital domain.

Direct-conversion (Homodyne / Zero-IF) receivers

Set $\omega_{LO}=\omega_{RF}$ so the channel lands directly at baseband (IF = 0). Because phase information is lost with a single mixer, direct-conversion uses quadrature (I/Q) downconversion: multiply by both $\cos\omega_0 t$ and $\sin\omega_0 t$.

LNA cos ω₀t LPF I sin ω₀t LPF Q • no image filter needed → LNA need not drive 50 Ω • channel select = LPF • far fewer mixing spurs
Zero-IF I/Q receiver: no image-reject filter, channel selection by on-chip LPF, fewer spurs. (Slide 9)
Why zero-IF is attractive for ICs
No image-rejection filter (so the LNA needn't drive a 50 Ω off-chip filter), channel selection by integrated low-pass filters, and dramatically fewer mixing spurs → ideal for monolithic integration.
…but two nasty issues appear at DC
1. LO leakage & self-mixing: the LO is at the RF frequency, so it leaks (through substrate / mixer ports) back to the antenna (re-radiation) and the leaked LO mixes with itself → a DC term at the output.
2. DC offsets: self-mixing plus device mismatch produce a static DC offset that sits right on top of the (now baseband) desired signal.
Antenna BPF LNA BPF RF IF DC offset LO LO self-mixing external LO leakage (re-radiation)
LO leakage paths: external leakage re-radiates from the antenna; internal LO self-mixing produces a DC offset. (Slide 9)

DC offset removal & I/Q mismatch

Three ways to kill the DC offset

  • High-pass filtering / AC coupling — a series capacitor $C_1$ + switch blocks DC. Simple, but it also notches out low-frequency signal content (bad for modulations with energy near DC).
  • Analog offset storage — sample the offset on a capacitor during an idle period, then subtract.
  • Digital offset storage — measure and cancel the offset digitally during a TDMA burst guard interval.
LNA cos ω₀t LPF C₁ S₁ dc offset 仍然會飄???the dc offset still drifts???dc オフセットはまだドリフトする??? t Offset Cancellation TDMA Burst 用數位的方式記錄發射時的直流準位,等到有訊號時扣掉它(digital offset storage)record the DC level digitally while transmitting, then subtract it once a signal arrives (digital offset storage)送信中の直流レベルをディジタルで記録しておき、信号が来たらそれを差し引く(digital offset storage)
Slides:AC 耦合AC couplingAC 結合(series $C_1$ + 開關 $S_1$)把 DC 擋掉——但開關切換與溫漂使 (a series $C_1$ plus switch $S_1$) blocks the DC — but switching transients and thermal drift mean the (直列 $C_1$ とスイッチ $S_1$)で DC を阻止する——しかしスイッチング過渡と温度ドリフトのためにdc offset 仍然會飄dc offset still driftsdc オフセットは依然としてドリフトする,這就是黃色問號的意思;, which is what the yellow question mark means; 。これが黄色い疑問符の意味である; digital offset storage 則在 TDMA burst 前的 offset-cancellation 區間量好直流準位,進入 burst 後用數位減掉。 instead measures the DC level during the offset-cancellation interval before a TDMA burst, then subtracts it digitally once the burst begins. は、TDMA バーストの前のオフセット除去区間で直流レベルを測定しておき、バーストが始まったらディジタルで差し引く。

I/Q mismatch

Quadrature downconversion needs the I and Q paths to be identical except for a perfect 90° phase difference. Real circuits have gain error and phase error between the paths:

  • Amplitude imbalance → the constellation stretches into a rectangle (I and Q scaled differently).
  • Phase imbalance → the constellation shears into a parallelogram (axes no longer orthogonal).
V_RF V_LO 90° LPF I LPF Q Phase and Gain Error Phase and Gain Error Phase and Gain Error Q I Amplitude imbalance Q I Phase imbalance
I/Q mismatch(slides):$V_{LO}$ 經 90° 分相、兩路 mixer + LPF,I/Q mismatch (slides): $V_{LO}$ passes through a 90° splitter into two mixer + LPF paths, and I/Q ミスマッチ(スライド):$V_{LO}$ が 90° 分配器を通って 2 系統の mixer + LPF に入り、每一段都會貢獻 phase 與 gain errorevery one of those stages contributes phase and gain errorその各段が位相誤差と利得誤差を生む。右兩圖:藍 = 理想方形星座;. The two figures on the right: blue = the ideal square constellation; 。右側の 2 つの図:青 = 理想的な正方形コンスタレーション; 橘 = amplitude imbalanceorange = amplitude imbalance橙 = 振幅アンバランス(I/Q 增益不同 → 拉成長方形);(unequal I/Q gain → stretched into a rectangle); (I/Q の利得が異なる → 長方形に引き伸ばされる); 粉 = phase imbalancepink = phase imbalance桃 = 位相アンバランス(兩軸不再正交 → 剪切成平行四邊形)。(the two axes are no longer orthogonal → sheared into a parallelogram).(2 軸が直交しなくなる → 平行四辺形にせん断される)。

Both rotate/distort the symbol constellation, raising the error-vector magnitude (EVM) and bit-error rate. I/Q mismatch is also what limits the image rejection of Hartley/Weaver receivers (next).

Even-order distortion (IP2)

In a direct-conversion receiver, even-order nonlinearity is uniquely dangerous. Two close interferers at $\omega_1,\omega_2$ produce a low-frequency beat at $\omega_1-\omega_2$ through second-order distortion. If this falls in the baseband and feeds through the mixer (via finite LO–RF isolation), it lands right on the desired channel. This effect is quantified by the second-order intercept point (IP2) — a spec that barely matters in heterodyne but is critical in zero-IF.

ω Desired Channel Interferers LNA cos ω_LO t Feedthrough ω 0 IM2 beat 疊在訊號上the IM2 beat lands on top of the signalIM2 ビートが信号に重なる This effect is quantified by the “IP2.” 用 differential 設計震盪器,減少 DC 含量(slides 註記)use a differential oscillator design to reduce the DC content (as annotated on the slides)発振器を差動構成にして DC 成分を減らす(スライドの注記)
Slide 11:兩個靠近的 interferers 經偶次失真產生 $\omega_1-\omega_2$ 的低頻 beat,藉 mixer 的有限 LO–RF 隔離 Slide 11: two closely spaced interferers produce a low-frequency beat at $\omega_1-\omega_2$ through even-order distortion, which then passes スライド 11:近接した 2 つの干渉波が偶数次歪みによって $\omega_1-\omega_2$ の低周波ビートを生じ、これが feedthrough 直通到輸出——在 zero-IF 裡正好疊在基頻訊號上。量化指標 = straight to the output via the mixer's finite LO–RF isolation — and in zero-IF it lands exactly on top of the baseband signal. The figure of merit is ミキサの有限な LO–RF 分離を通って出力へ直達する——zero-IF ではちょうどベースバンド信号の上に重なる。定量的な指標は IP2;對策之一是用 differential(差動)電路抑制偶次項與 DC 含量。; one remedy is a differential circuit, which suppresses the even-order terms and the DC content.;対策の一つは差動回路であり、偶数次項と DC 成分を抑制する。

Flicker (1/f) noise

Because the signal sits at baseband in zero-IF / low-IF receivers, the transistors' 1/f noise lands directly on it. The flicker PSD and total integrated noise over a band $[f_1,f_3]$ with corner $f_2$:

Flicker PSD and integrated noise $$S_{1/f}(f) = \frac{K}{W L C_{ox}}\,\frac{1}{f}$$ $$P_{n,total} = \frac{K}{WLC_{ox}}\ln\!\frac{f_2}{f_1} + (f_3-f_2)\,S_{th} = \Big(f_2\ln\tfrac{f_2}{f_1} + (f_3-f_2)\Big)S_{th}$$
這兩條式子怎麼來、怎麼用?(點擊展開)Where do these two expressions come from, and how are they used? (click to expand)この 2 つの式はどこから来て、どう使うのか?(クリックで展開)

1. $S_{1/f}$ 模型怎麼來的1. Where the $S_{1/f}$ model comes from1. $S_{1/f}$ モデルの由来

物理來源是 Si–SiO$_2$ 介面的缺陷(dangling bonds、interface traps)隨機捕捉/釋放通道載子;每個 trap 時間常數不同,大量疊加後 PSD 近似 $1/f$。這是 empirical model,折算到 gate 端的電壓 PSD,各因子的意義:The physical origin is defects at the Si–SiO$_2$ interface (dangling bonds, interface traps) randomly capturing and releasing channel carriers; each trap has a different time constant, and superposing a great many of them gives a PSD that is approximately $1/f$. It is an empirical model; referred to the gate as a voltage PSD, the factors mean:物理的な起源は Si–SiO$_2$ 界面の欠陥(ダングリングボンド、界面トラップ)がチャネルのキャリアをランダムに捕獲・放出することにある;トラップごとに時定数が異なり、多数を重ね合わせると PSD はほぼ $1/f$ になる。これは経験モデルであり、ゲート換算の電圧 PSD として表したときの各因子の意味は次のとおり:

  • $K$:製程常數,約 $10^{-25}\,\text{V}^2\!\cdot\!\text{F}$ 量級;PMOS 通常比 NMOS 小(載子離介面較遠)→ baseband 常用 PMOS input pair。: a process constant, of order $10^{-25}\,\text{V}^2\!\cdot\!\text{F}$; it is usually smaller for PMOS than NMOS (the carriers are further from the interface) → baseband stages commonly use a PMOS input pair.:プロセス定数で、$10^{-25}\,\text{V}^2\!\cdot\!\text{F}$ の程度;通常 PMOS のほうが NMOS より小さい(キャリアが界面から遠いため)→ ベースバンド段では PMOS 入力対がよく使われる。
  • $WL$ 在分母$WL$ in the denominator分母の $WL$:面積越大 = 對更多 trap 取平均,起伏互相抵消 → 雜訊降低。設計上最直接的武器:: a larger area averages over more traps, so the fluctuations cancel one another → lower noise. The most direct weapon available to the designer: :面積が大きいほど多くのトラップを平均化することになり、揺らぎが互いに打ち消し合う → 雑音が下がる。設計者にとって最も直接的な武器: 加大面積increase the area面積を大きくする.
  • $C_{ox}$ 在分母$C_{ox}$ in the denominator分母の $C_{ox}$:同樣的電荷起伏 $\Delta Q$,gate 電容越大,$\Delta V = \Delta Q/C$ 越小。: for the same charge fluctuation $\Delta Q$, a larger gate capacitance means a smaller $\Delta V = \Delta Q/C$.:同じ電荷揺らぎ $\Delta Q$ に対し、ゲート容量が大きいほど $\Delta V = \Delta Q/C$ は小さくなる。
  • $1/f$:往 DC 積分會發散,所以積分下限 $f_1$ 必須有界(上圖:由 PLL 頻寬決定)。: integrating down to DC diverges, so the lower integration limit $f_1$ must be bounded (in the figure above, it is set by the PLL bandwidth).:DC まで積分すると発散するので、積分の下限 $f_1$ は有界でなければならない(上図では PLL の帯域幅で決まる)。

2. $P_{n,total}$ 的推導2. Deriving $P_{n,total}$2. $P_{n,total}$ の導出

$f_2$ 是 $f_2$ is the $f_2$ は 1/f corner frequency——flicker PSD 與熱雜訊地板 $S_{th}$($\approx 4kT\gamma/g_m$,gate-referred)相等的頻率:— the frequency at which the flicker PSD equals the thermal noise floor $S_{th}$ ($\approx 4kT\gamma/g_m$, gate-referred):——フリッカ PSD が熱雑音フロア $S_{th}$($\approx 4kT\gamma/g_m$、ゲート換算)と等しくなる周波数:

$$\frac{K}{WLC_{ox}}\cdot\frac{1}{f_2} = S_{th}\;\Rightarrow\;\frac{K}{WLC_{ox}} = f_2\,S_{th}$$

總雜訊 = 兩段 PSD 各自積分再相加:$f_1\!\to\!f_2$ 積 $\dfrac{K}{WLC_{ox}f}$ 得 $\dfrac{K}{WLC_{ox}}\ln\dfrac{f_2}{f_1}$;$f_2\!\to\!f_3$ 積平坦的 $S_{th}$ 得 $(f_3-f_2)S_{th}$。把 $K/(WLC_{ox})=f_2 S_{th}$ 代入第一項,就得到右邊全部用 $S_{th}$ 表示的形式——Total noise = the integral of each PSD segment, added together: integrating $\dfrac{K}{WLC_{ox}f}$ from $f_1\!\to\!f_2$ gives $\dfrac{K}{WLC_{ox}}\ln\dfrac{f_2}{f_1}$; integrating the flat $S_{th}$ from $f_2\!\to\!f_3$ gives $(f_3-f_2)S_{th}$. Substituting $K/(WLC_{ox})=f_2 S_{th}$ into the first term yields the right-hand form expressed entirely in $S_{th}$ —総雑音 = 2 つの PSD 区間をそれぞれ積分して足し合わせたもの:$f_1\!\to\!f_2$ で $\dfrac{K}{WLC_{ox}f}$ を積分すると $\dfrac{K}{WLC_{ox}}\ln\dfrac{f_2}{f_1}$、$f_2\!\to\!f_3$ で平坦な $S_{th}$ を積分すると $(f_3-f_2)S_{th}$ となる。第 1 項に $K/(WLC_{ox})=f_2 S_{th}$ を代入すると、右辺がすべて $S_{th}$ で表された形になる——$K, W, L, C_{ox}$ 全部消掉,只要知道 corner $f_2$ 和頻帶邊界就能算$K, W, L, C_{ox}$ all cancel out, and all you need are the corner $f_2$ and the band edges$K, W, L, C_{ox}$ がすべて消え、コーナー $f_2$ と帯域端さえ分かれば計算できる.

3. 怎麼用:flicker noise penalty3. How to use it: the flicker-noise penalty3. 使い方:フリッカ雑音ペナルティ

用途是回答「比起只有 thermal noise,flicker 多吃幾 dB」:Its purpose is to answer “how many dB worse is this than thermal noise alone?”:用途は「熱雑音だけの場合と比べて何 dB 悪化するか」に答えることである:

$$\text{penalty} = 10\log\frac{f_2\ln(f_2/f_1) + (f_3-f_2)}{f_3-f_1}$$

數值感受(取 $f_2 = 200\,\text{kHz}$、$f_1 = 100\,\text{Hz}$,同 A feel for the numbers (taking $f_2 = 200\,\text{kHz}$ and $f_1 = 100\,\text{Hz}$, same 数値感覚($f_2 = 200\,\text{kHz}$、$f_1 = 100\,\text{Hz}$ とし、同じ S.2):

系統Systemシステム $f_3$ penalty 意義Meaning意味
窄頻(GSM)Narrowband (GSM)狭帯域(GSM) 200 kHz ≈ 8.8 dB 整段頻帶泡在 1/f 區,致命The whole band is submerged in the 1/f region — fatal帯域全体が 1/f 領域に浸かっており、致命的
寬頻(LTE/WLAN)Wideband (LTE/WLAN)広帯域(LTE/WLAN) 10 MHz ≈ 0.5 dB 只有前 2% 頻帶受害,幾乎無感Only the first 2% of the band suffers — barely noticeable帯域の最初の 2% しか影響を受けず、ほとんど気にならない

($f_3 < f_2$ 時整個頻帶都在 corner 以下:第二項取 0、第一項上限改 $f_3$。)(When $f_3 < f_2$ the entire band lies below the corner: take the second term as 0 and change the upper limit of the first term to $f_3$.)($f_3 < f_2$ のときは帯域全体がコーナーより下にある:第 2 項を 0 とし、第 1 項の上限を $f_3$ に変える。)

設計含義Design implications設計上の意味合い
zero-IF 把訊號搬到 DC 附近,flicker 直接疊在訊號上;可行性本質上是「channel 頻寬 ÷ flicker corner」的比值問題——窄頻系統(GSM)避用 zero-IF 而走 low-IF,寬頻系統(WLAN/LTE)可接受。電路對策:加大輸入對面積($WL\uparrow\Rightarrow f_2\downarrow$)、用 PMOS、chopping/auto-zero 把 1/f 調制走。 Zero-IF moves the signal down to near DC, so flicker noise lands directly on top of it; feasibility is essentially a question of the ratio “channel bandwidth ÷ flicker corner” — narrowband systems (GSM) avoid zero-IF and go low-IF, while wideband systems (WLAN/LTE) can live with it. Circuit remedies: enlarge the input-pair area ($WL\uparrow\Rightarrow f_2\downarrow$), use PMOS, and chop or auto-zero the 1/f away. zero-IF は信号を DC 付近まで下ろすため、フリッカ雑音が信号に直接重なる;成立するかどうかは本質的に「チャネル帯域幅 ÷ フリッカコーナー」の比の問題である——狭帯域システム(GSM)は zero-IF を避けて low-IF を採り、広帯域システム(WLAN/LTE)なら許容できる。回路的な対策:入力対の面積を大きくする($WL\uparrow\Rightarrow f_2\downarrow$)、PMOS を使う、チョッピングやオートゼロで 1/f を変調して除く。
f S₁/f = K/(WLCₒₓ) · 1/f S_th f₁ f₂ f₃ f₁ = PLL 頻寬f₁ = PLL bandwidthf₁ = PLL 帯域幅 PLL 讓這邊不會變成無限大(積分下限 f₁ 有界,ln(f₂/f₁) 不發散)The PLL keeps this from becoming infinite (the lower integration limit f₁ is bounded, so ln(f₂/f₁) does not diverge)PLL のおかげでここが無限大にならない(積分下限 f₁ が有界なので ln(f₂/f₁) は発散しない) S₁/f S_th f₁ 270 Hz f_u 100 kHz f₂ 積分結果(黃色面積 = flicker 貢獻)The integral (the yellow area = the flicker contribution)積分の結果(黄色の面積 = フリッカの寄与) flicker noise 越小越好the lower the flicker noise the betterフリッカ雑音は小さいほどよい
Slide 11:左 — flicker PSD 以 $1/f$ 下滑,在 corner $f_2$ 處與熱雜訊地板 $S_{th}$ 相等;積分下限 $f_1$ 由 Slide 11: left — the flicker PSD falls as $1/f$ and equals the thermal noise floor $S_{th}$ at the corner $f_2$; the lower integration limit $f_1$ is set by the スライド 11:左 — フリッカ PSD は $1/f$ で下がり、コーナー $f_2$ で熱雑音フロア $S_{th}$ と等しくなる;積分下限 $f_1$ は PLL 頻寬PLL bandwidthPLL 帯域幅(GSM 例:270 Hz)決定,所以 $\ln(f_2/f_1)$ 不會發散。右 — 總雜訊功率 = 黃色面積(flicker)+ 地板 × 頻寬;flicker corner 越低越好。(270 Hz in the GSM example), so $\ln(f_2/f_1)$ does not diverge. Right — total noise power = the yellow area (flicker) + floor × bandwidth; the lower the flicker corner the better.(GSM の例では 270 Hz)で決まるので、$\ln(f_2/f_1)$ は発散しない。右 — 総雑音電力 = 黄色の面積(フリッカ)+ フロア × 帯域幅;フリッカコーナーは低いほどよい。

Larger devices ($WL$) lower the flicker corner $f_2$. This is why zero-IF baseband stages use large input transistors and chopping/auto-zero techniques. (Slide 11)

Image-reject receivers — the 90° phase-shift idea

Since the image and the signal lie on opposite sides of the LO, they can be distinguished by their sign of frequency offset. The trick: a network that imparts a +90° shift to one sideband and −90° to the other, so that on recombination one sideband adds and the other cancels.

Shifting a modulated signal by 90° is the Hilbert transform: it multiplies positive frequencies by $-j$ and negative frequencies by $+j$. For a narrowband signal this is realized by an RC–CR network (one output through RC = lag, the other through CR = lead; they differ by 90° at the center frequency).

RC–CR all-pass 90° network V_in R V_out1 C C V_out2 R out1, out2 differ by 90° at ω₀
RC–CR network: one output leads, the other lags, giving a 90° split at the center frequency. (Slide 12)

Hartley architecture

The Hartley image-reject receiver mixes the RF with quadrature LOs ($\sin\omega_{LO}t$ and $\cos\omega_{LO}t$), low-pass filters each path, applies a 90° shift to one path, and sums. The desired channel adds constructively while the image cancels.

RF in sin ω_LO t LPF A 90° C cos ω_LO t LPF B + IF Output desired → adds image → cancels
Hartley: quadrature mix → LPF → 90° shift on one branch → sum. The RC–CR realizes the 90°. (Slide 13)
Hartley architecture spectral walkthrough
Hartley 頻譜推演:RF input 的 desired channel 與 image 對稱分佈在 $\pm\omega_{LO}$ 兩側。Hartley spectral walkthrough: the desired channel and the image of the RF input sit symmetrically on either side of $\pm\omega_{LO}$.Hartley のスペクトル追跡:RF 入力の希望チャネルとイメージは $\pm\omega_{LO}$ の両側に対称に位置する。sin 路徑The sin pathsin 経路與 LO 的脈衝對($\mp j/2$ 於 $\pm\omega_{LO}$)做卷積得 $X_A(\omega)$——正負頻率各帶 $+j/2$、$-j/2$,desired 與 image convolves with the LO's impulse pair ($\mp j/2$ at $\pm\omega_{LO}$) to give $X_A(\omega)$ — the positive and negative frequencies carry $+j/2$ and $-j/2$, and the desired signal and the image have は LO のインパルス対($\pm\omega_{LO}$ における $\mp j/2$)と畳み込まれて $X_A(\omega)$ となる——正の周波数と負の周波数がそれぞれ $+j/2$、$-j/2$ を持ち、希望波とイメージは 符號相反opposite signs符号が逆になる;cos 路徑The cos pathcos 経路($+1/2$ 於 $\pm\omega_{LO}$)得 $X_B(\omega)$,兩者($+1/2$ at $\pm\omega_{LO}$) gives $X_B(\omega)$, in which the two have ($\pm\omega_{LO}$ における $+1/2$)からは $X_B(\omega)$ が得られ、そこでは両者が 符號相同the same sign同じ符号になる。對 $X_A$ 施加 90° 移相(Hilbert:正頻 $\times(-j)$、負頻 $\times(+j)$)得 $X_C(\omega)$:desired 翻成與 $X_B$ 同號、image 翻成反號 → 相加後 . Applying a 90° phase shift to $X_A$ (a Hilbert transform: positive frequencies $\times(-j)$, negative frequencies $\times(+j)$) gives $X_C(\omega)$: the desired signal flips to the same sign as $X_B$ while the image flips to the opposite sign → on adding them, 。$X_A$ に 90° 移相(ヒルベルト変換:正の周波数 $\times(-j)$、負の周波数 $\times(+j)$)を施すと $X_C(\omega)$ が得られる:希望波は $X_B$ と同符号に、イメージは逆符号に反転する → 加算すると desired 相加the desired signal adds希望波は加算され and image 相消the image cancelsイメージは打ち消される,IF output 不含 image。, so the IF output contains no image.ので、IF 出力にイメージは含まれない。
為何 90° 移相會讓正負頻率「有的維持、有的反向」?(點擊展開)Why does a 90° phase shift leave some frequencies alone and invert others? (click to expand)なぜ 90° 移相は、ある周波数をそのままにし、別の周波数を反転させるのか?(クリックで展開)

這裡有兩個獨立的機制疊在一起,分開看就清楚了。Two independent mechanisms are superimposed here; separate them and it becomes clear.ここでは 2 つの独立した機構が重なっている。分けて見れば明快になる。

1. 實體的「90° 移相器」對正負頻率本來就必須做相反的事(realizability)1. A physical “90° phase shifter” must inherently do opposite things to positive and negative frequencies (realizability)1. 物理的な「90° 移相器」は、正の周波数と負の周波数に対して本質的に逆のことをせざるを得ない(実現可能性)

輸入輸出都是實數訊號的網路(RC–CR 就是),其 transfer function 必滿足 conjugate symmetry:$H(-\omega)=H^*(\omega)$。若 $|H|=1$,phase 必須是 $\omega$ 的A network with real input and output (an RC–CR is one) must have a transfer function satisfying conjugate symmetry: $H(-\omega)=H^*(\omega)$. If $|H|=1$, the phase must be an 入力も出力も実数信号であるネットワーク(RC–CR がまさにそれ)は、伝達関数が共役対称 $H(-\omega)=H^*(\omega)$ を満たさなければならない。$|H|=1$ ならば、位相は $\omega$ の奇函數odd function奇関数——所以不可能對所有頻率都乘 $-j$,只能是: of $\omega$ — so it cannot multiply every frequency by $-j$; it can only be:でなければならない——したがってすべての周波数に $-j$ を掛けることはできず、取りうるのは次の形だけである:

$$H(\omega)=-j\,\mathrm{sgn}(\omega)\quad{\scriptstyle\textcolor{gray}{\text{正頻率}\times e^{-j90^\circ}=-j,\ \text{負頻率}\times e^{+j90^\circ}=+j\ \text{——這就是 Hilbert transform}}}$$ $$H(\omega)=-j\,\mathrm{sgn}(\omega)\quad{\scriptstyle\textcolor{gray}{\text{positive frequencies}\times e^{-j90^\circ}=-j,\ \text{negative frequencies}\times e^{+j90^\circ}=+j\ \text{— this is the Hilbert transform}}}$$ $$H(\omega)=-j\,\mathrm{sgn}(\omega)\quad{\scriptstyle\textcolor{gray}{\text{正の周波数}\times e^{-j90^\circ}=-j,\ \text{負の周波数}\times e^{+j90^\circ}=+j\ \text{——これがヒルベルト変換である}}}$$

直觀說法:實數訊號是一對共軛的反向旋轉 phasor($e^{+j\omega t}$ 與 $e^{-j\omega t}$),要讓輸出仍是實數,兩支必須保持共軛——逆時針那支轉 $-90^\circ$,順時針那支就得轉 $+90^\circ$。Intuitively: a real signal is a conjugate pair of counter-rotating phasors ($e^{+j\omega t}$ and $e^{-j\omega t}$), and for the output to stay real the pair must remain conjugate — rotate the anticlockwise one by $-90^\circ$ and the clockwise one must rotate by $+90^\circ$.直観的には:実数信号は互いに共役な逆回転フェーザの対($e^{+j\omega t}$ と $e^{-j\omega t}$)であり、出力を実数に保つにはこの対が共役のままでなければならない——反時計回りの成分を $-90^\circ$ 回すなら、時計回りの成分は $+90^\circ$ 回す必要がある。

2. 「有的維持、有的反向」不是移相器區別對待,而是輸入自己帶的符號不同2. “Some kept, some inverted” is not the phase shifter discriminating — the input itself arrives with different signs2. 「そのまま/反転」は移相器が区別しているのではなく、入力自身が異なる符号を帯びて入ってくるためである

移相器對所有成分一視同仁地乘 $-j\,\mathrm{sgn}(\omega)$;差別來自 $X_A(\omega)$ 裡 desired 和 image 進來時就帶著The phase shifter multiplies every component by $-j\,\mathrm{sgn}(\omega)$ without distinction; the difference is that inside $X_A(\omega)$ the desired signal and the image already arrive with 移相器はすべての成分に区別なく $-j\,\mathrm{sgn}(\omega)$ を掛ける;違いは、$X_A(\omega)$ の中で希望波とイメージがすでに相反的虛數符號opposite imaginary signs逆の虚数符号。原因:它們在 RF 端位於 $\omega_{LO}$ 的兩側,而 $\sin$ 的頻譜是奇虛函數($+j/2$ 在 $-\omega_{LO}$、$-j/2$ 在 $+\omega_{LO}$),卷積之後一個落在 $+j/2$、一個落在 $-j/2$——這正是上圖 $X_A$ 標 $\pm j/2$ 的意思。接著做乘法,關鍵是 $(\pm j)\times(\pm j)=\pm 1$:. The reason: at RF they lie on opposite sides of $\omega_{LO}$, and the spectrum of $\sin$ is an odd imaginary function ($+j/2$ at $-\omega_{LO}$, $-j/2$ at $+\omega_{LO}$), so after convolution one lands on $+j/2$ and the other on $-j/2$ — which is exactly what the $\pm j/2$ labels on $X_A$ in the figure above mean. Then comes the multiplication, and the key is that $(\pm j)\times(\pm j)=\pm 1$:を帯びて入ってくることにある。理由:RF では両者が $\omega_{LO}$ の反対側に位置し、$\sin$ のスペクトルは奇な純虚関数($-\omega_{LO}$ で $+j/2$、$+\omega_{LO}$ で $-j/2$)なので、畳み込みの後に一方は $+j/2$、他方は $-j/2$ に落ちる——これが上図で $X_A$ に $\pm j/2$ と記されている意味である。次に掛け算を行うが、鍵は $(\pm j)\times(\pm j)=\pm 1$ である:

在 $+\omega_{IF}$At $+\omega_{IF}$$+\omega_{IF}$ において 在 $-\omega_{IF}$At $-\omega_{IF}$$-\omega_{IF}$ において 結果Result結果
desired $(+\tfrac{j}{2})(-j)=+\tfrac12$ $(-\tfrac{j}{2})(+j)=+\tfrac12$ 正的實偶函數 → 與 cos 路徑 $X_B$ a positive real even function → with the cos path $X_B$ it 正の実偶関数 → cos 経路の $X_B$ と 同號the same sign同符号
image $(-\tfrac{j}{2})(-j)=-\tfrac12$ $(+\tfrac{j}{2})(+j)=-\tfrac12$ 負的實偶函數 → 與 $X_B$ a negative real even function → with $X_B$ it 負の実偶関数 → $X_B$ と 反號opposite signs逆符号

時域驗證Time-domain check時間領域での確認:sin 路徑出來 desired 是 $-\tfrac{A}{2}\sin\omega_{IF}t$、image 是 $+\tfrac{B}{2}\sin\omega_{IF}t$(符號相反)。整條路再延遲 $90^\circ$:$\sin(\omega_{IF}t-90^\circ)=-\cos\omega_{IF}t$,所以 desired → $+\tfrac{A}{2}\cos$、image → $-\tfrac{B}{2}\cos$;而 cos 路徑兩者都是 $+\cos$。相加後 desired 加倍、image 歸零。: out of the sin path, the desired signal is $-\tfrac{A}{2}\sin\omega_{IF}t$ and the image is $+\tfrac{B}{2}\sin\omega_{IF}t$ (opposite signs). Delaying the whole path by a further $90^\circ$: $\sin(\omega_{IF}t-90^\circ)=-\cos\omega_{IF}t$, so the desired signal becomes $+\tfrac{A}{2}\cos$ and the image $-\tfrac{B}{2}\cos$; on the cos path both are $+\cos$. Adding them doubles the desired signal and nulls the image.:sin 経路から出てくる希望波は $-\tfrac{A}{2}\sin\omega_{IF}t$、イメージは $+\tfrac{B}{2}\sin\omega_{IF}t$(符号が逆)。この経路をさらに $90^\circ$ 遅らせると $\sin(\omega_{IF}t-90^\circ)=-\cos\omega_{IF}t$ なので、希望波は $+\tfrac{A}{2}\cos$、イメージは $-\tfrac{B}{2}\cos$ になる;一方 cos 経路では両者とも $+\cos$ である。加算すると希望波は 2 倍になり、イメージは 0 になる。

一句話總結In one sentence一言でまとめると
90° 移相器只做一件事($-j\,\mathrm{sgn}(\omega)$);「維持或反向」是因為 desired 與 image 經過 sin mixer 後本來就帶相反的 $\pm j$,同一個旋轉把它們分別轉到實軸的正、負兩端。 The 90° phase shifter does only one thing ($-j\,\mathrm{sgn}(\omega)$); “kept or inverted” arises because, after the sin mixer, the desired signal and the image already carry opposite $\pm j$, so the very same rotation takes them to opposite ends of the real axis. 90° 移相器がしていることは 1 つだけ($-j\,\mathrm{sgn}(\omega)$)である;「そのまま/反転」が生じるのは、sin ミキサを通った後の希望波とイメージがもともと逆の $\pm j$ を帯びているからであり、同じ回転が両者を実軸の正端と負端へ振り分けるためである。
📝 HW12 — Derive the Image-Rejection Ratio (IRR)

With a fractional gain mismatch ε and a phase mismatch θ between the I and Q paths, the image is no longer perfectly cancelled. Derive the image-rejection ratio and show it equals:

IRR (image-to-signal leakage) $$\text{IRR} = \frac{I}{S} = \frac{1 - 2(1+\varepsilon)\cos\theta + (1+\varepsilon)^2}{1 + 2(1+\varepsilon)\cos\theta + (1+\varepsilon)^2}$$
RF Input A LO sin ωLOt A LO(1+ε) cos(ωLOt+θ) ε = 增益誤差,θ = 相位誤差ε = gain error, θ = phase errorε = 利得誤差、θ = 位相誤差 LPF A 90° C LPF B + IF Output hard to do(類比域難做準)hard to do (tough to get right in the analogue domain)hard to do(アナログ領域では精度を出しにくい)
Hartley 加上實際誤差(slides):LO 兩路不再完美正交——增益差 $(1+\varepsilon)$、相位差 $\theta$;黃色的 Hartley with real-world errors (slides): the two LO paths are no longer perfectly in quadrature — there is a gain difference $(1+\varepsilon)$ and a phase difference $\theta$; the yellow 実際の誤差を加えた Hartley(スライド):LO の 2 経路はもはや完全な直交ではない——利得差 $(1+\varepsilon)$ と位相差 $\theta$ がある;黄色の 90° 移相器90° phase shifter90° 移相器 and 電壓加法器voltage adder電圧加算器是類比域最難做準的兩塊(hard to do),誤差直接決定 IRR。 are the two hardest blocks to get right in the analogue domain (hard to do), and their errors set the IRR directly. はアナログ領域で最も精度を出しにくい 2 つのブロックであり(hard to do)、その誤差が IRR を直接決めてしまう。
IRR [dB] Θ [°] 30 40 50 60 70 80 90 0.001 0.01 0.1 1 e = 0.1 dB e = 0.03 dB e = 0.01 dB e = 0.003 dB e = 0.001 dB e = 20 log(1+ε) [dB],Θ = θ in degreese = 20 log(1+ε) [dB], Θ = θ in degreese = 20 log(1+ε) [dB]、Θ = θ(度)
IRR 對 LO 相位誤差 Θ(log 軸)與增益誤差 e 的關係(slides):每條曲線左端平坦段由IRR as a function of LO phase error Θ (log axis) and gain error e (slides): the flat left-hand portion of each curve is set by the IRR と LO の位相誤差 Θ(対数軸)および利得誤差 e の関係(スライド):各曲線の左側の平坦部は增益誤差gain error利得誤差決定($\text{IRR}\approx\varepsilon^2/4$),右端共同上揚段由($\text{IRR}\approx\varepsilon^2/4$), while the common rising right-hand portion is set by the で決まり($\text{IRR}\approx\varepsilon^2/4$)、右側で共通に立ち上がる部分は相位誤差phase error位相誤差決定($\approx\theta^2/4$);60 dB 需要 e ≤ 0.01 dB 且 Θ ≤ 0.1° ——類比域幾乎做不到,典型只有 30–35 dB。($\approx\theta^2/4$); 60 dB requires e ≤ 0.01 dB and Θ ≤ 0.1° — all but unachievable in the analogue domain, where 30–35 dB is typical.で決まる($\approx\theta^2/4$);60 dB を得るには e ≤ 0.01 dB かつ Θ ≤ 0.1° が必要——アナログ領域ではほぼ実現不可能で、典型的には 30–35 dB にとどまる。
點此展開推導 · Show derivationClick to expand the derivation · Show derivationクリックして導出を表示 · Show derivation

Sketch of the derivation. Let the desired and image phasors propagate through the two paths. Path gains are $1$ and $(1+\varepsilon)$; the intended 90° has an error $\theta$ so one phasor is $(1+\varepsilon)e^{j(\pi/2+\theta)}$ instead of $e^{j\pi/2}$. At the summing node:

  • Desired sideband: the two contributions are designed to add → amplitude $\propto \big|1 + (1+\varepsilon)e^{j\theta}\big|$.
  • Image sideband: designed to cancel → residual amplitude $\propto \big|1 - (1+\varepsilon)e^{j\theta}\big|$.

Taking the power ratio $|1-(1+\varepsilon)e^{j\theta}|^2 / |1+(1+\varepsilon)e^{j\theta}|^2$ and expanding $|a\pm b e^{j\theta}|^2 = a^2 + b^2 \pm 2ab\cos\theta$ gives the boxed formula. For small errors it reduces to the handy approximation:

$$\text{IRR} \approx \frac{\varepsilon^2 + \theta^2}{4}\quad\Rightarrow\quad \text{IRR}_{dB}\approx 10\log_{10}\!\frac{\varepsilon^2+\theta^2}{4}$$

Numbers: 60 dB rejection needs a phase mismatch of only ~0.1° (and matching amplitude). Typical realized IRR is 30–35 dB.

Other Hartley weaknesses: (1) voltage addition at the output node; (2) amplitude imbalance in the RC–CR network at frequencies away from center; (3) the $1/(RC)$ corner drifts with process and temperature. (Slide 14)

Weaver architecture

The Weaver receiver replaces the troublesome 90° RC–CR phase shifter with a second quadrature mixing stage. Two complex mixing operations (at $\omega_1$ then $\omega_2$) separate the sidebands purely by mixing, with the outputs summed in the current domain.

RF in sin ω₁t LPF sin ω₂t C cos ω₁t LPF cos ω₂t D IF Output • no RC dependence • sum in current domain • must avoid secondary image
Weaver: two quadrature mixing stages (ω₁ then ω₂); no RC phase shifter, but the second mix introduces a secondary image to guard against. (Slide 15)
Weaver architecture spectral walkthrough (slide 15)
Slides 原圖(p.15):下半是 Weaver 的頻譜推演 — 上路 $X_A(\omega)$(sin 路徑,正負頻率各帶 $\pm j$)與下路 $X_B(\omega)$(cos 路徑)分別與第二級 LO 的脈衝對($\mp j/2$ 於 $\pm\omega_2$;$+1/2$ 於 $\pm\omega_2$)做卷積,得 $X_C, X_D$;相減後 desired 相加、image 相消。藍字註記:secondary image(2nd image signal still)仍須避開;多出一個本地振盪源、吃功耗。Original slide figure (p.15): the lower half is the Weaver spectral walkthrough — the upper path $X_A(\omega)$ (the sin path, with $\pm j$ on the positive and negative frequencies) and the lower path $X_B(\omega)$ (the cos path) are each convolved with the second-stage LO's impulse pair ($\mp j/2$ at $\pm\omega_2$; $+1/2$ at $\pm\omega_2$) to give $X_C, X_D$; on subtracting, the desired signal adds and the image cancels. The blue annotations: the secondary image (2nd image signal still) must still be avoided; and it costs an extra local oscillator and more power.スライド原図(p.15):下半分は Weaver のスペクトル追跡 — 上側経路 $X_A(\omega)$(sin 経路、正負の周波数がそれぞれ $\pm j$ を持つ)と下側経路 $X_B(\omega)$(cos 経路)を、それぞれ第 2 段 LO のインパルス対($\pm\omega_2$ における $\mp j/2$;$\pm\omega_2$ における $+1/2$)と畳み込んで $X_C, X_D$ を得る;減算すると希望波は加算され、イメージは打ち消される。青字の注記:二次イメージ(2nd image signal still)は依然として避けなければならない;局部発振源が 1 つ増え、消費電力も増える。
FeatureHartleyWeaver
90° realized byRC–CR network2nd quadrature mixer
Main weaknessRC mismatch / drift limits IRRSecondary image from 2nd mix
Output summationVoltage domainCurrent domain (better)
Process sensitivityHigh (depends on RC)Low (no RC)

Low-IF receivers — the compromise

The low-IF architecture downconverts to a small, non-zero IF (a few channel bandwidths) instead of exactly DC, using quadrature mixing and digitizing at that low IF.

preselect VCO/LO (I,Q) BPF BPF ADC ADC digital image reject radio freq low intermediate freq baseband image rejection done in baseband/DSP
Low-IF: quadrature mix to a small IF, digitize, reject the image digitally in baseband. (Slide 16)
Why low-IF is "the compromise"
  • Better than heterodyne — no bulky analog IF block / off-chip IF filter.
  • Better than homodyne (zero-IF) — the signal is offset from DC, so it is largely immune to DC offset, LO self-mixing leakage, and 1/f flicker noise.
  • The cost — the image now sits only a couple of channels away, so it demands high image rejection performed in baseband / DSP (digital complex filtering or polyphase networks).

6.3 Transmitter Architectures

Basic structures — the SSB (quadrature) up-mixer

A transmitter modulates baseband I/Q data onto a carrier. The core building block is the single-sideband (SSB) mixer: I and Q baseband streams multiply $\cos\omega_{LO}t$ and $\sin\omega_{LO}t$ respectively and are summed, producing one sideband (the modulated signal) and cancelling the other.

Serial/Parallel Converter Baseband Data sin ω_LO t cos ω_LO t + Modulated Signal SSB Mixer
Quadrature up-conversion (SSB mixer): I·sin + Q·cos → single-sideband modulated RF. (Slide 17)

Baseband pulse shaping (slide 17): I/Q symbols are read from ROM look-up tables → DAC → LPF before the mixer, producing band-limited (e.g., raised-cosine) pulses that meet the spectral mask.

Direct-conversion transmitter

The direct-conversion TX is the inverse of the zero-IF receiver: baseband I/Q modulate quadrature carriers directly at the final RF, sum, then feed the PA → matching network → duplexer → antenna.

Baseband I cos ω_c t Baseband Q sin ω_c t + PA Matching Network Duplexer
Direct-conversion TX: baseband I/Q → quadrature up-mix → sum → PA → matching → duplexer → antenna. (Slide 18)

Carrier feedthrough & LO pulling (and the offset-LO fix)

Two transmitter-specific problems
Carrier feedthrough: any DC offset in the baseband I/Q multiplies the carrier and appears as an un-modulated carrier leakage in the RF output spectrum.

LO pulling (injection pulling): the PA output is a large signal at the same frequency as the VCO. It couples back and pulls the VCO frequency, corrupting the modulation.

Solution: the offset LO

Generate the carrier from two VCOs at different frequencies ($\omega_1$ and $\omega_2$) and mix them so the carrier is $\omega_1+\omega_2$ — neither VCO runs at the PA frequency, so the PA cannot pull them.

Pulling (problem) VCO + BPF PA PA pulls the VCO Offset LO (solution) VCO₁ ω₁ VCO₂ ω₂ BPF ω₁+ω₂ carrier = ω₁+ω₂; no VCO at PA freq
Left: PA output pulls a same-frequency VCO. Right: offset-LO scheme synthesizes the carrier as ω₁+ω₂ so no oscillator runs at the PA frequency. (Slide 18)

Two-step transmitter

The two-step (heterodyne) TX upconverts the I/Q to a first IF, band-pass filters, then mixes up to the final RF. Because the quadrature up-conversion happens at the lower IF (where matching the I/Q paths is easier), I/Q mismatch is reduced. The penalty is that the second mix creates an image that needs filtering.

I sin ω₁t Q cos ω₁t + BPF cos ω₂t BPF PA ω₁+ω₂ + I/Q matched at low IF − image after 2nd mix → needs image filtering / SSB mixers to relax it
Two-step TX: I/Q up-mix at IF (good matching) → BPF → up-mix to RF → BPF → PA. Reduces I/Q mismatch but requires image filtering. (Slide 19)

6.4 Case Studies & Summary

Case studies

  • Tri-band WCDMA transceiver (900 MHz / 2 GHz / 2.5 GHz): multi-band LNAs, shared synthesizer with clock distribution, dual-PA, gain-step LO/PGA control — illustrates how the architectures above are combined and reused across bands. (Slide 20)
  • 802.11a/b/g transceiver (dual-band 2.4 GHz & 5 GHz): direct-conversion (zero-IF) I/Q paths, shared fractional-N PLL with a ÷2 quadrature generator, separate 2 GHz / 5 GHz LNAs and PAs, baseband PGAs, RSSI, TX detector. A textbook modern CMOS zero-IF radio. (Slide 21)

Receiver architecture comparison

ArchitectureImage handlingKey strengthsKey problems
Heterodyne (single)RF image-reject filterMature, robust, high dynamic rangeOff-chip filters; image vs channel-select trade-off
Dual-conversionHigh IF₁ then low IF₂Breaks the IF trade-off; excellent selectivitySecondary image; more LOs/filters; power & cost
Zero-IF (homodyne)None (I/Q baseband)Fully integrable; no image filter; few spursDC offset, LO self-mixing, 1/f noise, I/Q & IP2
Image-reject (Hartley)90° RC–CR + sumCancels image electronicallyRC drift limits IRR (~30–35 dB)
Image-reject (Weaver)2nd quadrature mixNo RC dependence; current-domain sumSecondary image to avoid
Low-IFDigital/baseband image rejectAvoids DC offset & 1/f; integrableHigh baseband image-rejection demand
RX — the big picture

Everything is a fight with the image and with where the signal sits relative to DC. Heterodyne pushes the image away with filters; zero-IF removes the image but inherits DC/flicker problems; low-IF and image-reject mixers are the modern compromises.

TX — the big picture

Quadrature SSB up-conversion is the core. Direct-conversion is simplest but suffers carrier feedthrough (baseband DC) and LO pulling (PA→VCO). Offset-LO and two-step architectures trade complexity for cleaner spectra and lower I/Q mismatch.

Practice Problems — Cross-Section Synthesis

Four main problems, each chaining several sections of this chapter · Detailed answers included

SynthesisCross-Section Synthesis

Each problem deliberately chains material from several sections — frequency planning (6.2), zero-IF impairments (6.2), image-reject architectures (6.2), and transmitter design (6.3) — the way a real architecture review does.

S.1 Hard
Heterodyne frequency plan — image vs channel-select, dual conversion, secondary image
Synthesis of 6.2: heterodyne, the image problem, the IF trade-off, dual downconversion, HW11.
System Receiver at $f_{RF} = 900\,\text{MHz}$, channel bandwidth B = 200 kHz, high-side LO
RF front-end BPF: 60 dB attenuation at ±140 MHz offset, but only 25 dB at ±20 MHz
Practical band-pass channel-select filters are limited to $Q \approx 50$

(a) With $f_{IF} = 70\,\text{MHz}$: give $f_{LO}$ and the image frequency, and check the image rejection. Repeat for $f_{IF} = 10\,\text{MHz}$. What trade-off do the two cases demonstrate?

(b) Can the channel be selected directly at 70 MHz? (Compute the required filter Q.)

(c) Fix the problem with a second conversion to $f_{IF2} = 10.7\,\text{MHz}$ (low-side LO₂). Give $f_{LO2}$ and the required Q at IF₂.

(d) Find the secondary image: which frequency at the first IF also converts to 10.7 MHz, and which RF frequencies land there? Which filter must remove them?

Answer

(a) 高 IF 與低 IF 的 image 位置與抑制(a) Where the image lands and how well it is rejected, at high and low IF(a) 高い IF と低い IF におけるイメージの位置と抑圧量

已知/目標Given / required既知/目標:$f_{RF}=900\,\text{MHz}$、$B=200\,\text{kHz}$、high-side LO($f_{LO}>f_{RF}$);前端 BPF 在 ±140 MHz offset 給 60 dB、在 ±20 MHz 只給 25 dB。分別對 $f_{IF}=70\,\text{MHz}$ 與 $10\,\text{MHz}$ 求 $f_{LO}$、image frequency,並檢查 image rejection。: $f_{RF}=900\,\text{MHz}$, $B=200\,\text{kHz}$, high-side LO ($f_{LO}>f_{RF}$); the front-end BPF gives 60 dB at ±140 MHz offset but only 25 dB at ±20 MHz. For $f_{IF}=70\,\text{MHz}$ and $10\,\text{MHz}$ respectively, find $f_{LO}$ and the image frequency, and check the image rejection.:$f_{RF}=900\,\text{MHz}$、$B=200\,\text{kHz}$、high-side LO($f_{LO}>f_{RF}$);フロントエンド BPF は ±140 MHz オフセットで 60 dB、±20 MHz では 25 dB しか与えない。$f_{IF}=70\,\text{MHz}$ と $10\,\text{MHz}$ のそれぞれについて $f_{LO}$ とイメージ周波数を求め、イメージ抑圧を確認せよ。

公式與來源Formulas and sources公式と出典:high-side injection 時 $f_{LO}=f_{RF}+f_{IF}$。Mixer 對任何滿足 $|f-f_{LO}|=f_{IF}$ 的輸入一視同仁,所以 image 是 desired 對 $f_{LO}$ 的鏡像,永遠落在離 desired 恰好 $2f_{IF}$ 處(6.2 the image problem):: for high-side injection $f_{LO}=f_{RF}+f_{IF}$. The mixer treats any input satisfying $|f-f_{LO}|=f_{IF}$ identically, so the image is the mirror of the desired signal about $f_{LO}$ and always lands exactly $2f_{IF}$ from it (6.2, the image problem)::high-side injection では $f_{LO}=f_{RF}+f_{IF}$。ミキサは $|f-f_{LO}|=f_{IF}$ を満たす入力をすべて同じように扱うので、イメージは希望波の $f_{LO}$ に関する鏡像であり、常に希望波からちょうど $2f_{IF}$ だけ離れた位置に落ちる(6.2 the image problem):

$$f_{im}=f_{LO}+f_{IF}=f_{RF}+2f_{IF}$$

逐步代入Substituting step by step順に代入($f_{IF}=70\,\text{MHz}$):($f_{IF}=70\,\text{MHz}$):($f_{IF}=70\,\text{MHz}$):

  • $f_{LO}=900+70=970\,\text{MHz}$
  • $f_{im}=900+2\times 70=1040\,\text{MHz}$
  • image offset $=1040-900=140\,\text{MHz}$ → 正好落在 BPF 的 ±140 MHz 規格點,吃到完整的 60 dB rejection ✓image offset $=1040-900=140\,\text{MHz}$ → exactly on the BPF's ±140 MHz specification point, so it gets the full 60 dB of rejection ✓image offset $=1040-900=140\,\text{MHz}$ → ちょうど BPF の ±140 MHz 規格点に当たり、60 dB の抑圧をフルに受けられる ✓

逐步代入Substituting step by step順に代入($f_{IF}=10\,\text{MHz}$):($f_{IF}=10\,\text{MHz}$):($f_{IF}=10\,\text{MHz}$):

  • $f_{LO}=900+10=910\,\text{MHz}$、$f_{im}=900+2\times 10=920\,\text{MHz}$$f_{LO}=900+10=910\,\text{MHz}$, $f_{im}=900+2\times 10=920\,\text{MHz}$$f_{LO}=900+10=910\,\text{MHz}$、$f_{im}=900+2\times 10=920\,\text{MHz}$
  • image offset 只剩 20 MHz → BPF 只給 25 dB。典型 out-of-band blocker 可以比 desired 強 60–80 dB,25 dB 完全不夠。the image offset is only 20 MHz → the BPF gives just 25 dB. A typical out-of-band blocker can be 60–80 dB stronger than the desired signal, so 25 dB is nowhere near enough.image offset は 20 MHz しか残らない → BPF は 25 dB しか与えない。典型的な帯域外ブロッカーは希望波より 60–80 dB 強くなりうるので、25 dB ではまったく足りない。
High IF → image far away, easy image rejection, but hard channel selection. Low IF → easy channel selection, hopeless image. This is THE heterodyne trade-off (6.2).

物理詮釋Physical interpretation物理的な解釈:image 落點由 $2f_{IF}$ 唯一決定,與前端 filter 的 roll-off 無關——選 IF 就是在替 image「選位置」,這是 frequency planning 的第一個自由度。: where the image lands is determined solely by $2f_{IF}$, independently of the front-end filter's roll-off — choosing the IF is choosing where to put the image, and that is the first degree of freedom in frequency planning.:イメージの落ちる位置は $2f_{IF}$ だけで一意に決まり、フロントエンドフィルタのロールオフとは無関係である——IF を選ぶことはイメージの「置き場所」を選ぶことであり、これが周波数プランニングにおける第一の自由度である。

(b) 直接在 70 MHz 做 channel selection?(b) Can channel selection be done directly at 70 MHz?(b) 70 MHz で直接チャネル選択はできるか?

已知/目標Given / required既知/目標:channel bandwidth $B=200\,\text{kHz}$、filter 中心頻率 $f_0=f_{IF}=70\,\text{MHz}$、可實作 $Q\lesssim 50$。求所需 Q。: channel bandwidth $B=200\,\text{kHz}$, filter centre frequency $f_0=f_{IF}=70\,\text{MHz}$, realizable $Q\lesssim 50$. Find the Q required.:チャネル帯域幅 $B=200\,\text{kHz}$、フィルタ中心周波数 $f_0=f_{IF}=70\,\text{MHz}$、実現可能な $Q\lesssim 50$。必要な Q を求めよ。

公式與來源Formulas and sources公式と出典:band-pass filter 的 loaded Q 定義為中心頻率除以 −3 dB 頻寬:$Q=f_0/B$;要恰好罩住一個 channel,頻寬就取 $B=200\,\text{kHz}$。: the loaded Q of a band-pass filter is the centre frequency divided by the −3 dB bandwidth, $Q=f_0/B$; to cover exactly one channel, take $B=200\,\text{kHz}$.:帯域通過フィルタの負荷 Q は中心周波数を −3 dB 帯域幅で割ったもの、すなわち $Q=f_0/B$ である;チャネルをちょうど 1 つ覆うには $B=200\,\text{kHz}$ とする。

逐步代入Substituting step by step順に代入:

$$Q=\frac{f_{IF}}{B}=\frac{70\times 10^6}{200\times 10^3}=350\gg 50$$
所需 Q = 350,超出可實作上限(≈50)整整 7 倍 → 70 MHz 直接選 channel 不可行The required Q = 350, a full 7× beyond the realizable ceiling (≈50) → selecting the channel directly at 70 MHz is not feasible必要な Q = 350 で、実現可能な上限(≈50)を丸 7 倍も超える → 70 MHz で直接チャネル選択するのは実現不可能

物理詮釋Physical interpretation物理的な解釈:fractional bandwidth $B/f_0\approx 0.29\%$ 太窄——filter 難度由相對頻寬決定,這就是為什麼 channel selection 一定要把訊號再往低頻搬。: the fractional bandwidth $B/f_0\approx 0.29\%$ is far too narrow — filter difficulty is governed by relative bandwidth, which is why channel selection always requires moving the signal further down in frequency.:比帯域幅 $B/f_0\approx 0.29\%$ は狭すぎる——フィルタの難易度は相対帯域幅で決まる。これこそが、チャネル選択では必ず信号をさらに低い周波数へ移さなければならない理由である。

(c) Dual conversion 到 $f_{IF2}=10.7$ MHz(c) Dual conversion to $f_{IF2}=10.7$ MHz(c) $f_{IF2}=10.7$ MHz へのダブルコンバージョン

已知/目標Given / required既知/目標:$f_{IF1}=70\,\text{MHz}$、low-side LO₂($f_{LO2}<f_{IF1}$)、$f_{IF2}=10.7\,\text{MHz}$。求 $f_{LO2}$ 與 IF₂ 的所需 Q。: $f_{IF1}=70\,\text{MHz}$, low-side LO₂ ($f_{LO2}<f_{IF1}$), $f_{IF2}=10.7\,\text{MHz}$. Find $f_{LO2}$ and the Q required at IF₂.:$f_{IF1}=70\,\text{MHz}$、low-side LO₂($f_{LO2}<f_{IF1}$)、$f_{IF2}=10.7\,\text{MHz}$。$f_{LO2}$ と IF₂ で必要な Q を求めよ。

公式與來源Formulas and sources公式と出典:low-side injection:$f_{LO2}=f_{IF1}-f_{IF2}$;Q 與 (b) 同樣用 $Q=f_0/B$。: low-side injection: $f_{LO2}=f_{IF1}-f_{IF2}$; the Q follows $Q=f_0/B$ as in (b).:low-side injection:$f_{LO2}=f_{IF1}-f_{IF2}$;Q は (b) と同じく $Q=f_0/B$ による。

逐步代入Substituting step by step順に代入:

  • $f_{LO2}=70-10.7=\mathbf{59.3\,MHz}$
  • $Q=\dfrac{10.7\times 10^6}{200\times 10^3}=53.5\approx 50$ ✓(剛好落在可實作邊緣)$Q=\dfrac{10.7\times 10^6}{200\times 10^3}=53.5\approx 50$ ✓ (right on the edge of what is realizable)$Q=\dfrac{10.7\times 10^6}{200\times 10^3}=53.5\approx 50$ ✓(実現可能なぎりぎりの境界)
fLO2 = 59.3 MHz;所需 Q ≈ 53.5 ≈ 50 → 可實作。Dual conversion 拆解了 (a) 的 trade-off:IF₁ 高負責 image rejection,IF₂ 低負責 selectivity。 = 59.3 MHz; the required Q ≈ 53.5 ≈ 50 → realizable. Dual conversion untangles the trade-off of part (a): a high IF₁ handles image rejection while a low IF₂ handles selectivity. = 59.3 MHz;必要な Q ≈ 53.5 ≈ 50 → 実現可能。ダブルコンバージョンは (a) のトレードオフを分解する:高い IF₁ がイメージ抑圧を担い、低い IF₂ が選択度を担う。

物理詮釋Physical interpretation物理的な解釈:一次降頻無法同時滿足「image 遠」與「Q 低」,兩段降頻讓兩個目標各自擁有專屬的 IF——這正是傳統 superheterodyne 接收機的標準解法。: a single downconversion cannot satisfy “image far away” and “low Q” at once; two stages give each goal its own dedicated IF — precisely the standard solution of the classical superheterodyne receiver.:一度の周波数変換では「イメージが遠い」と「Q が低い」を同時に満たせない。2 段に分ければそれぞれの目標に専用の IF を与えられる——これこそ古典的なスーパーヘテロダイン受信機の標準的な解である。

(d) Secondary image

已知/目標Given / required既知/目標:第二級 mixer($f_{LO2}=59.3\,\text{MHz}$)會把任何滿足 $|f-f_{LO2}|=f_{IF2}$ 的 IF₁ 頻率都轉到 10.7 MHz。找出 70 MHz 以外的解、它對應的 RF 頻率,以及該由哪個 filter 負責。: the second mixer ($f_{LO2}=59.3\,\text{MHz}$) converts any IF₁ frequency satisfying $|f-f_{LO2}|=f_{IF2}$ down to 10.7 MHz. Find the solution other than 70 MHz, the RF frequency it corresponds to, and which filter should deal with it.:第 2 段ミキサ($f_{LO2}=59.3\,\text{MHz}$)は $|f-f_{LO2}|=f_{IF2}$ を満たす IF₁ 周波数をすべて 10.7 MHz へ変換する。70 MHz 以外の解と、それに対応する RF 周波数、そしてどのフィルタが担当すべきかを求めよ。

公式與來源Formulas and sources公式と出典:第二級的 image 是 desired IF₁ 對 LO₂ 的鏡像(與 (a) 同一個原理,套在第二級):$f_{im,IF1}=2f_{LO2}-f_{IF1}$。再反推第一級:第一個 mixer($f_{LO}=970\,\text{MHz}$)把 $|f_{RF}-f_{LO}|=f_{im,IF1}$ 的 RF 都搬到該處。: the second stage's image is the mirror of the desired IF₁ about LO₂ (the same principle as (a), applied to the second stage): $f_{im,IF1}=2f_{LO2}-f_{IF1}$. Then work back to the first stage: the first mixer ($f_{LO}=970\,\text{MHz}$) moves any RF satisfying $|f_{RF}-f_{LO}|=f_{im,IF1}$ to that point.:第 2 段のイメージは、希望 IF₁ の LO₂ に関する鏡像である((a) と同じ原理を第 2 段に適用したもの):$f_{im,IF1}=2f_{LO2}-f_{IF1}$。次に第 1 段へ遡ると、第 1 ミキサ($f_{LO}=970\,\text{MHz}$)は $|f_{RF}-f_{LO}|=f_{im,IF1}$ を満たす RF をすべてそこへ移す。

逐步代入Substituting step by step順に代入:

  • 解 $|f-59.3|=10.7$:$f=59.3+10.7=70\,\text{MHz}$(desired)或 $f=59.3-10.7=\mathbf{48.6\,MHz}$(即 $2\times 59.3-70=48.6$,70 MHz 對 LO₂ 的鏡像)Solving $|f-59.3|=10.7$: $f=59.3+10.7=70\,\text{MHz}$ (desired) or $f=59.3-10.7=\mathbf{48.6\,MHz}$ (i.e. $2\times 59.3-70=48.6$, the mirror of 70 MHz about LO₂)$|f-59.3|=10.7$ を解くと:$f=59.3+10.7=70\,\text{MHz}$(希望波)または $f=59.3-10.7=\mathbf{48.6\,MHz}$(すなわち $2\times 59.3-70=48.6$ で、70 MHz の LO₂ に関する鏡像)
  • 反推 RF:$|f_{RF}-970|=48.6$ → $f_{RF}=970-48.6=921.4\,\text{MHz}$ 或 $f_{RF}=970+48.6=1018.6\,\text{MHz}$Working back to RF: $|f_{RF}-970|=48.6$ → $f_{RF}=970-48.6=921.4\,\text{MHz}$ or $f_{RF}=970+48.6=1018.6\,\text{MHz}$RF へ遡ると:$|f_{RF}-970|=48.6$ → $f_{RF}=970-48.6=921.4\,\text{MHz}$ または $f_{RF}=970+48.6=1018.6\,\text{MHz}$
Secondary image at RF: 921.4 MHz and 1018.6 MHz

物理詮釋Physical interpretation物理的な解釈:921.4 MHz 離 desired 的 900 MHz 只有 21.4 MHz——RF preselector 幾乎擋不住(同 (a) 低 IF 情形的 25 dB 等級)。但它在 IF₁ 上是 48.6 MHz 對 70 MHz,fractional offset $(70-48.6)/70\approx 30\%$,一個普通的 : 921.4 MHz is only 21.4 MHz from the desired 900 MHz — the RF preselector can barely block it (the same ~25 dB level as the low-IF case in (a)). But at IF₁ it is 48.6 MHz against 70 MHz, a fractional offset of $(70-48.6)/70\approx 30\%$, so an ordinary :921.4 MHz は希望波の 900 MHz からわずか 21.4 MHz しか離れていない——RF プリセレクタではほとんど阻止できない((a) の低 IF の場合と同じ 25 dB 程度)。しかし IF₁ では 70 MHz に対して 48.6 MHz であり、比オフセットは $(70-48.6)/70\approx 30\%$ になるので、ごく普通の IF₁ band-pass filter(中心 70 MHz)就能輕鬆壓掉。這就是 dual-conversion receiver 兩個 mixer 之間永遠要放 BPF 的原因(HW11)。(centred at 70 MHz) suppresses it easily. This is why a dual-conversion receiver always has a BPF between its two mixers (HW11).(中心 70 MHz)で容易に抑圧できる。これが、ダブルコンバージョン受信機では 2 つのミキサの間に必ず BPF を置く理由である(HW11)。

S.2 Hard
Zero-IF impairment budget — DC offset, IP2, flicker, I/Q mismatch
Synthesis of 6.2: direct conversion, DC offset removal, IP2, flicker noise, I/Q mismatch + HW12 formula.
Zero-IF receiver (GSM-like) B = 200 kHz; desired signal at sensitivity −99 dBm; voltage gain LNA→mixer output 30 dB; 50 Ω reference
LO: 0 dBm; LO-to-LNA-input isolation 60 dB
Two −35 dBm interferers near the carrier; IM2 beat must stay ≥ 9 dB below the desired signal
Flicker corner $f_2 = 200\,\text{kHz}$; offset-cancellation HPF corner $f_1 = 100\,\text{Hz}$
I/Q paths: gain error 0.2 dB, phase error 2°

(a) Estimate the DC offset from LO self-mixing at the mixer output (in dBm and mV), compare it with the desired signal level there, and pick the right offset-removal method for a TDMA system.

(b) What IIP2 must the front end have?

(c) Using the flicker formula of 6.2, compute the noise penalty relative to thermal-only for this 200 kHz channel, then for a 10 MHz LTE channel. What does the comparison teach?

(d) Compute the image-rejection ratio from the HW12 small-error formula. Is it enough for QPSK (≈ −20 dB needed)? For 64-QAM (≈ −40 dB)?

Answer

(a) LO self-mixing 造成的 DC offset(a) The DC offset caused by LO self-mixing(a) LO セルフミキシングによる DC オフセット

已知/目標Given / required既知/目標:$P_{LO}=0\,\text{dBm}$、LO→LNA-input isolation 60 dB、LNA→mixer-output 電壓增益 30 dB、desired 在 sensitivity $-99\,\text{dBm}$、50 Ω 系統。求 mixer 輸出端的 DC offset(dBm 與 mV)、與 desired 的差距,並為 TDMA 系統選 offset-removal 方法。: $P_{LO}=0\,\text{dBm}$, LO→LNA-input isolation 60 dB, LNA→mixer-output voltage gain 30 dB, the desired signal at the sensitivity of $-99\,\text{dBm}$, a 50 Ω system. Find the DC offset at the mixer output (in dBm and mV), how far above the desired signal it is, and choose an offset-removal method for a TDMA system.:$P_{LO}=0\,\text{dBm}$、LO→LNA 入力の分離 60 dB、LNA→ミキサ出力の電圧利得 30 dB、希望波は感度 $-99\,\text{dBm}$、50 Ω 系。ミキサ出力での DC オフセット(dBm と mV)、希望波との差、そして TDMA 系に適したオフセット除去法を求めよ。

公式與來源Formulas and sources公式と出典:zero-IF 的 LO 經有限 isolation 漏到 LNA 輸入,被放大後回到 mixer 與 LO 本尊相乘(self-mixing),$\cos\omega t\cdot\cos\omega t = \tfrac12(1+\cos 2\omega t)$ 產生 DC 項(6.2 direct conversion)。量級估計:$P_{DC}\approx P_{LO}-\text{isolation}+G$。功率轉電壓用 $V_{rms}=\sqrt{P\cdot R}$。: in zero-IF the LO leaks through the finite isolation to the LNA input, is amplified, and returns to the mixer where it multiplies with the LO itself (self-mixing); $\cos\omega t\cdot\cos\omega t = \tfrac12(1+\cos 2\omega t)$ produces a DC term (6.2, direct conversion). Order-of-magnitude estimate: $P_{DC}\approx P_{LO}-\text{isolation}+G$. Converting power to voltage uses $V_{rms}=\sqrt{P\cdot R}$.:zero-IF では LO が有限の分離を通って LNA 入力へ漏れ、増幅されてミキサへ戻り、LO 本体と掛け合わされる(セルフミキシング)。$\cos\omega t\cdot\cos\omega t = \tfrac12(1+\cos 2\omega t)$ が DC 項を生む(6.2 direct conversion)。オーダー評価は $P_{DC}\approx P_{LO}-\text{isolation}+G$。電力から電圧への換算には $V_{rms}=\sqrt{P\cdot R}$ を使う。

逐步代入Substituting step by step順に代入:

  • LNA 輸入端的 LO leakage:$0-60=-60\,\text{dBm}$LO leakage at the LNA input: $0-60=-60\,\text{dBm}$LNA 入力での LO 漏れ:$0-60=-60\,\text{dBm}$
  • 經 30 dB 增益到 mixer 輸出再 self-mix:$-60+30=\mathbf{-30\,dBm}$after 30 dB of gain to the mixer output, then self-mixing: $-60+30=\mathbf{-30\,dBm}$30 dB の利得を経てミキサ出力に達し、そこでセルフミキシング:$-60+30=\mathbf{-30\,dBm}$
  • dBm→W:$-30\,\text{dBm}=10^{-30/10}\,\text{mW}=10^{-3}\,\text{mW}=1\,\mu\text{W}$dBm→W: $-30\,\text{dBm}=10^{-30/10}\,\text{mW}=10^{-3}\,\text{mW}=1\,\mu\text{W}$dBm→W:$-30\,\text{dBm}=10^{-30/10}\,\text{mW}=10^{-3}\,\text{mW}=1\,\mu\text{W}$
  • $V_{rms}=\sqrt{10^{-6}\times 50}=\sqrt{5\times 10^{-5}}=\mathbf{7.1\,mV}$(DC offset)$V_{rms}=\sqrt{10^{-6}\times 50}=\sqrt{5\times 10^{-5}}=\mathbf{7.1\,mV}$ (the DC offset)$V_{rms}=\sqrt{10^{-6}\times 50}=\sqrt{5\times 10^{-5}}=\mathbf{7.1\,mV}$(DC オフセット)
  • desired 在同一節點:$-99+30=-69\,\text{dBm}=10^{-6.9}\,\text{mW}=1.26\times 10^{-10}\,\text{W}$ → $V_{rms}=\sqrt{1.26\times 10^{-10}\times 50}=\mathbf{79\,\mu V}$the desired signal at the same node: $-99+30=-69\,\text{dBm}=10^{-6.9}\,\text{mW}=1.26\times 10^{-10}\,\text{W}$ → $V_{rms}=\sqrt{1.26\times 10^{-10}\times 50}=\mathbf{79\,\mu V}$同じノードでの希望波:$-99+30=-69\,\text{dBm}=10^{-6.9}\,\text{mW}=1.26\times 10^{-10}\,\text{W}$ → $V_{rms}=\sqrt{1.26\times 10^{-10}\times 50}=\mathbf{79\,\mu V}$
  • 差距:$-30-(-69)=\mathbf{39\,dB}$(電壓比 $7.1\,\text{mV}/79\,\mu\text{V}\approx 89$ 倍)the gap: $-30-(-69)=\mathbf{39\,dB}$ (a voltage ratio of $7.1\,\text{mV}/79\,\mu\text{V}\approx 89$)差:$-30-(-69)=\mathbf{39\,dB}$(電圧比で $7.1\,\text{mV}/79\,\mu\text{V}\approx 89$ 倍)
For TDMA: digital offset storage — measure and subtract the offset during the idle/guard interval of each burst (6.2). AC coupling alone would notch signal energy near DC.

物理詮釋Physical interpretation物理的な解釈:一個比訊號大 39 dB 的 DC 項會先把 baseband amplifier/ADC 的 dynamic range 吃光、甚至直接 saturate——訊號根本還沒被偵測到。TDMA 有現成的 idle/guard interval 可以量測並數位扣除 offset;若只靠 AC coupling,corner 必須極低(訊號能量集中在 DC 附近)導致 settling 太慢,跟不上 burst 切換。: a DC term 39 dB larger than the signal will consume the entire dynamic range of the baseband amplifier and ADC, or saturate them outright — the signal is never even detected. TDMA has a ready-made idle/guard interval in which the offset can be measured and subtracted digitally; relying on AC coupling alone would require an extremely low corner (the signal energy is concentrated near DC), making the settling far too slow to keep up with burst switching.:信号より 39 dB 大きい DC 項は、ベースバンド増幅器と ADC のダイナミックレンジを食い尽くすか、そのまま飽和させてしまう——信号は検出すらされない。TDMA には、オフセットを測定してディジタルで差し引くための idle/guard 区間が最初から用意されている;AC 結合だけに頼ると、信号のエネルギーが DC 付近に集中しているためコーナーを極端に低くせざるを得ず、整定が遅すぎてバースト切り替えに追随できない。

(b) IIP2 需求(b) The IIP2 requirement(b) IIP2 の要求値

已知/目標Given / required既知/目標:兩個 $-35\,\text{dBm}$ 的 interferer,其 IM2 beat(差頻項 $f_1-f_2$,在 zero-IF 直接落進 baseband)必須比 desired($-99\,\text{dBm}$)再低 9 dB。求前端所需 IIP2。: two interferers at $-35\,\text{dBm}$, whose IM2 beat (the difference term $f_1-f_2$, which in zero-IF lands directly in baseband) must be a further 9 dB below the desired signal ($-99\,\text{dBm}$). Find the IIP2 required of the front end.:$-35\,\text{dBm}$ の干渉波が 2 つあり、その IM2 ビート(差の項 $f_1-f_2$ で、zero-IF では直接ベースバンドに落ちる)は希望波($-99\,\text{dBm}$)よりさらに 9 dB 低くなければならない。フロントエンドに必要な IIP2 を求めよ。

公式與來源Formulas and sources公式と出典:二階互調的 input-referred beat level(two-tone、等功率 $P_{in}$):: the input-referred beat level of second-order intermodulation (two tones of equal power $P_{in}$)::2 次相互変調の入力換算ビートレベル(等電力 $P_{in}$ の 2 トーン):

$$P_{IM2,in}=2P_{in}-IIP2$$

(每 1 dB 的輸入變化讓 IM2 變 2 dB,外插交點即 IIP2。)(every 1 dB change of input moves the IM2 by 2 dB, and the extrapolated intercept is the IIP2.)(入力が 1 dB 変わるごとに IM2 は 2 dB 動き、外挿した交点が IIP2 である。)

逐步代入Substituting step by step順に代入:

  • $P_{IM2,in}=2\times(-35)-IIP2=-70-IIP2$(dBm)$P_{IM2,in}=2\times(-35)-IIP2=-70-IIP2$ (dBm)$P_{IM2,in}=2\times(-35)-IIP2=-70-IIP2$(dBm)
  • 要求 $P_{IM2,in}\leq -99-9=-108\,\text{dBm}$requiring $P_{IM2,in}\leq -99-9=-108\,\text{dBm}$$P_{IM2,in}\leq -99-9=-108\,\text{dBm}$ が要求される
  • $-70-IIP2\leq -108$ → $IIP2\geq -70+108=+38\,\text{dBm}$
IIP2 ≥ −70 + 108 = +38 dBm — a spec that "barely matters in heterodyne but is critical in zero-IF" (6.2); met with differential mixers + calibration

物理詮釋Physical interpretation物理的な解釈:heterodyne 中 IM2 的 beat 落在 baseband、會被 IF filter 擋掉,所以 IIP2 幾乎不設規格;zero-IF 的「IF」就是 baseband,beat 直接疊在訊號上——+38 dBm 這種誇張的數字只能靠全差動 mixer(理論上消偶次項)加上校準達成。: in a heterodyne the IM2 beat lands in baseband and is blocked by the IF filter, so IIP2 is barely specified at all; in zero-IF the “IF” is baseband and the beat lands straight on top of the signal — an outlandish figure like +38 dBm can only be reached with a fully differential mixer (which cancels the even-order terms in theory) plus calibration.:ヘテロダインでは IM2 のビートがベースバンドに落ち、IF フィルタで阻止されるため IIP2 はほとんど規定されない;zero-IF では「IF」がベースバンドそのものなので、ビートが信号の上に直接重なる——+38 dBm というような極端な値は、完全差動ミキサ(理論上は偶数次項を打ち消す)と校正を併用してはじめて達成できる。

(c) Flicker noise penalty:GSM vs LTE(c) The flicker-noise penalty: GSM vs LTE(c) フリッカ雑音ペナルティ:GSM と LTE

已知/目標Given / required既知/目標:flicker corner $f_2=200\,\text{kHz}$、offset-cancellation HPF corner $f_1=100\,\text{Hz}$、channel 上緣 $f_3=$ 200 kHz(GSM)或 10 MHz(LTE)。求相對 thermal-only 的 noise penalty。: flicker corner $f_2=200\,\text{kHz}$, offset-cancellation HPF corner $f_1=100\,\text{Hz}$, upper channel edge $f_3=$ 200 kHz (GSM) or 10 MHz (LTE). Find the noise penalty relative to thermal-only.:フリッカコーナー $f_2=200\,\text{kHz}$、オフセット除去 HPF のコーナー $f_1=100\,\text{Hz}$、チャネル上端 $f_3=$ 200 kHz(GSM)または 10 MHz(LTE)。熱雑音のみの場合に対する雑音ペナルティを求めよ。

公式與來源Formulas and sources公式と出典:baseband noise PSD 在 $f<f_2$ 處為 $S_{th}\cdot f_2/f$(1/f 區),$f>f_2$ 處為 $S_{th}$(平坦區)。對 $f_1\to f_3$ 積分(6.2 flicker 公式):: the baseband noise PSD is $S_{th}\cdot f_2/f$ for $f<f_2$ (the 1/f region) and $S_{th}$ for $f>f_2$ (the flat region). Integrating from $f_1\to f_3$ (the 6.2 flicker formula)::ベースバンド雑音 PSD は $f<f_2$ で $S_{th}\cdot f_2/f$(1/f 領域)、$f>f_2$ で $S_{th}$(平坦領域)である。$f_1\to f_3$ で積分すると(6.2 のフリッカの式):

$$P_n=\int_{f_1}^{f_2}S_{th}\frac{f_2}{f}\,df+\int_{f_2}^{f_3}S_{th}\,df=\Big(f_2\ln\frac{f_2}{f_1}+(f_3-f_2)\Big)S_{th}$$

thermal-only 基準為 $(f_3-f_1)S_{th}$。the thermal-only reference is $(f_3-f_1)S_{th}$.熱雑音のみの基準は $(f_3-f_1)S_{th}$ である。

逐步代入Substituting step by step順に代入(GSM,$f_3=f_2=200\,\text{kHz}$,平坦區為零):(GSM, $f_3=f_2=200\,\text{kHz}$, so the flat region contributes nothing):(GSM、$f_3=f_2=200\,\text{kHz}$ なので平坦領域の寄与はゼロ):

  • $\ln(f_2/f_1)=\ln(200{,}000/100)=\ln(2000)=7.60$
  • $P_n=200\times 10^3\times 7.60\,S_{th}=1.52\times 10^6\,S_{th}$
  • thermal-only:$(200\times 10^3-100)\,S_{th}\approx 2\times 10^5\,S_{th}$thermal-only: $(200\times 10^3-100)\,S_{th}\approx 2\times 10^5\,S_{th}$thermal-only:$(200\times 10^3-100)\,S_{th}\approx 2\times 10^5\,S_{th}$
  • penalty $=10\log\dfrac{1.52\times 10^6}{2\times 10^5}=10\log(7.6)=\mathbf{8.8\,dB}$

逐步代入Substituting step by step順に代入(LTE,$f_3=10\,\text{MHz}$):(LTE, $f_3=10\,\text{MHz}$):(LTE、$f_3=10\,\text{MHz}$):

  • 1/f 區貢獻不變:$1.52\times 10^6\,S_{th}$;平坦區:$(10\times 10^6-0.2\times 10^6)\,S_{th}=9.8\times 10^6\,S_{th}$the 1/f contribution is unchanged: $1.52\times 10^6\,S_{th}$; the flat region: $(10\times 10^6-0.2\times 10^6)\,S_{th}=9.8\times 10^6\,S_{th}$1/f 領域の寄与は変わらず $1.52\times 10^6\,S_{th}$;平坦領域は $(10\times 10^6-0.2\times 10^6)\,S_{th}=9.8\times 10^6\,S_{th}$
  • $P_n=(1.52+9.8)\times 10^6\,S_{th}=1.13\times 10^7\,S_{th}$;thermal-only $\approx 10^7\,S_{th}$$P_n=(1.52+9.8)\times 10^6\,S_{th}=1.13\times 10^7\,S_{th}$; thermal-only $\approx 10^7\,S_{th}$$P_n=(1.52+9.8)\times 10^6\,S_{th}=1.13\times 10^7\,S_{th}$;thermal-only は $\approx 10^7\,S_{th}$
  • penalty $=10\log(1.13)=\mathbf{0.5\,dB}$
Flicker noise punishes narrowband zero-IF hard (8.8 dB) but wideband barely (0.5 dB) — why zero-IF won in WLAN/LTE first, and why GSM zero-IF needed large devices + chopping (6.2)

物理詮釋Physical interpretation物理的な解釈:1/f 雜訊的總能量固定集中在低頻;channel 越寬,被污染的比例越小。GSM 的 200 kHz 幾乎整段都泡在 1/f 區,LTE 的 10 MHz 只有前 2% 受害——zero-IF 的可行性本質上是「頻寬除以 flicker corner」的比值問題。: the total energy of 1/f noise is fixed and concentrated at low frequencies, so the wider the channel the smaller the fraction that is contaminated. Almost the whole of GSM's 200 kHz is submerged in the 1/f region, whereas only the first 2% of LTE's 10 MHz suffers — the feasibility of zero-IF is essentially a question of the ratio “bandwidth divided by flicker corner”.:1/f 雑音の総エネルギーは低周波に集中したまま一定なので、チャネルが広いほど汚染される割合は小さくなる。GSM の 200 kHz はほぼ全体が 1/f 領域に浸かっているのに対し、LTE の 10 MHz は最初の 2% しか影響を受けない——zero-IF が成立するかどうかは、本質的に「帯域幅をフリッカコーナーで割った比」の問題である。

(d) I/Q mismatch 的 IRR(d) IRR from I/Q mismatch(d) I/Q ミスマッチによる IRR

已知/目標Given / required既知/目標:gain error 0.2 dB、phase error 2°。用 HW12 small-error 公式求 image-rejection ratio,判斷 QPSK(需 ≈ −20 dB)與 64-QAM(需 ≈ −40 dB)。: a gain error of 0.2 dB and a phase error of 2°. Use the HW12 small-error formula to find the image-rejection ratio, and judge it against QPSK (needs ≈ −20 dB) and 64-QAM (needs ≈ −40 dB).:利得誤差 0.2 dB、位相誤差 2°。HW12 の微小誤差の式でイメージ抑圧比を求め、QPSK(≈ −20 dB 必要)と 64-QAM(≈ −40 dB 必要)に照らして判定せよ。

公式與來源Formulas and sources公式と出典:HW12 的 small-error 近似($\varepsilon$ 為線性 gain error、$\theta$ 為弧度制 phase error):: the HW12 small-error approximation ($\varepsilon$ the linear gain error, $\theta$ the phase error in radians)::HW12 の微小誤差近似($\varepsilon$ は線形の利得誤差、$\theta$ はラジアン単位の位相誤差):

$$\text{IRR}=\frac{P_{image}}{P_{signal}}\approx\frac{\varepsilon^2+\theta^2}{4}$$

逐步代入Substituting step by step順に代入:

  • dB→線性:$\varepsilon=10^{0.2/20}-1=1.0233-1=0.0233$ → $\varepsilon^2=5.4\times 10^{-4}$dB→linear: $\varepsilon=10^{0.2/20}-1=1.0233-1=0.0233$ → $\varepsilon^2=5.4\times 10^{-4}$dB→線形:$\varepsilon=10^{0.2/20}-1=1.0233-1=0.0233$ → $\varepsilon^2=5.4\times 10^{-4}$
  • 度→弧度:$\theta=2°\times\dfrac{\pi}{180}=0.0349\,\text{rad}$ → $\theta^2=1.22\times 10^{-3}$degrees→radians: $\theta=2°\times\dfrac{\pi}{180}=0.0349\,\text{rad}$ → $\theta^2=1.22\times 10^{-3}$度→ラジアン:$\theta=2°\times\dfrac{\pi}{180}=0.0349\,\text{rad}$ → $\theta^2=1.22\times 10^{-3}$
  • $\text{IRR}=\dfrac{5.4\times 10^{-4}+1.22\times 10^{-3}}{4}=\dfrac{1.76\times 10^{-3}}{4}=4.4\times 10^{-4}$
  • dB:$10\log(4.4\times 10^{-4})=\mathbf{-33.6\,dB}$in dB: $10\log(4.4\times 10^{-4})=\mathbf{-33.6\,dB}$dB 表示:$10\log(4.4\times 10^{-4})=\mathbf{-33.6\,dB}$
IRR = −33.6 dB:QPSK(≈ −20 dB)✓ 通過;64-QAM(≈ −40 dB)✗ 不足 6 dBIRR = −33.6 dB: QPSK (≈ −20 dB) ✓ passes; 64-QAM (≈ −40 dB) ✗ short by 6 dBIRR = −33.6 dB:QPSK(≈ −20 dB)✓ 合格;64-QAM(≈ −40 dB)✗ 6 dB 不足

物理詮釋Physical interpretation物理的な解釈:0.2 dB/2° 已是不錯的類比匹配,仍離 64-QAM 的需求差 6 dB 以上——而且本例 phase 項($1.22\times 10^{-3}$)貢獻是 gain 項的兩倍多。這就是現代 radio 一律在類比匹配之上再加 digital I/Q calibration 的原因。: 0.2 dB / 2° is already respectable analogue matching, yet it still falls more than 6 dB short of what 64-QAM needs — and in this example the phase term ($1.22\times 10^{-3}$) contributes more than twice as much as the gain term. This is why every modern radio adds digital I/Q calibration on top of analogue matching.:0.2 dB/2° はすでにかなり良いアナログ整合だが、それでも 64-QAM の要求には 6 dB 以上足りない——しかも本例では位相項($1.22\times 10^{-3}$)の寄与が利得項の 2 倍以上ある。これが、現代の無線機がアナログ整合の上にディジタル I/Q 校正を必ず重ねる理由である。

S.3 Hard
Hartley vs Weaver — IRR numbers, RC–CR drift, secondary image
Synthesis of 6.2: image-reject receivers, Hartley, HW12, Weaver, low-IF.

(a) A Hartley receiver has gain mismatch $\varepsilon = 0.3\,\text{dB}$ and phase mismatch $\theta = 3°$. Compute the IRR and compare with the "typical 30–35 dB" claim of 6.2.

(b) The RC–CR network's $1/RC$ corner drifts +20% with process. Show what happens to the two branch amplitudes and phases at the (unchanged) signal frequency, and the resulting IRR. Which error — amplitude or phase — kills Hartley?

(c) A Weaver receiver takes RF = 2.4 GHz with $\omega_1 = 2.3\,\text{GHz}$ (IF₁ = 100 MHz) and wants IF₂ = 20 MHz. Compare the two choices $\omega_2 = 120\,\text{MHz}$ vs $\omega_2 = 80\,\text{MHz}$: where is the secondary image at IF₁ in each case, and which choice can the inter-stage LPF protect?

(d) Summarize: why does Weaver beat Hartley on process sensitivity, and what limits both?

Answer

(a) Hartley 的 IRR(a) The IRR of a Hartley receiver(a) Hartley 受信機の IRR

已知/目標Given / required既知/目標:gain mismatch $\varepsilon=0.3\,\text{dB}$、phase mismatch $\theta=3°$。求 IRR 並與 6.2 的「typical 30–35 dB」對照。: gain mismatch $\varepsilon=0.3\,\text{dB}$, phase mismatch $\theta=3°$. Find the IRR and compare it with the “typical 30–35 dB” of 6.2.:利得ミスマッチ $\varepsilon=0.3\,\text{dB}$、位相ミスマッチ $\theta=3°$。IRR を求め、6.2 の「典型値 30–35 dB」と比較せよ。

公式與來源Formulas and sources公式と出典:HW12 small-error 公式($\varepsilon$ 取線性、$\theta$ 取弧度):: the HW12 small-error formula ($\varepsilon$ linear, $\theta$ in radians)::HW12 の微小誤差の式($\varepsilon$ は線形、$\theta$ はラジアン):

$$\text{IRR}=\frac{\varepsilon^2+\theta^2}{4}$$

逐步代入Substituting step by step順に代入:

  • dB→線性:$\varepsilon=10^{0.3/20}-1=1.0352-1=0.0352$ → $\varepsilon^2=1.24\times 10^{-3}$dB→linear: $\varepsilon=10^{0.3/20}-1=1.0352-1=0.0352$ → $\varepsilon^2=1.24\times 10^{-3}$dB→線形:$\varepsilon=10^{0.3/20}-1=1.0352-1=0.0352$ → $\varepsilon^2=1.24\times 10^{-3}$
  • 度→弧度:$\theta=3°\times\dfrac{\pi}{180}=0.0524\,\text{rad}$ → $\theta^2=2.74\times 10^{-3}$degrees→radians: $\theta=3°\times\dfrac{\pi}{180}=0.0524\,\text{rad}$ → $\theta^2=2.74\times 10^{-3}$度→ラジアン:$\theta=3°\times\dfrac{\pi}{180}=0.0524\,\text{rad}$ → $\theta^2=2.74\times 10^{-3}$
  • $\text{IRR}=\dfrac{1.24\times 10^{-3}+2.74\times 10^{-3}}{4}=\dfrac{3.98\times 10^{-3}}{4}=9.9\times 10^{-4}$
  • dB:$10\log(9.9\times 10^{-4})=\mathbf{-30\,dB}$in dB: $10\log(9.9\times 10^{-4})=\mathbf{-30\,dB}$dB 表示:$10\log(9.9\times 10^{-4})=\mathbf{-30\,dB}$
IRR = −30 dB — right at the bottom of the typical 30–35 dB range(6.2)✓IRR = −30 dB — right at the bottom of the typical 30–35 dB range (6.2) ✓IRR = −30 dB — 典型値 30–35 dB の下限にちょうど一致する(6.2)✓

物理詮釋Physical interpretation物理的な解釈:0.3 dB/3° 是未經校準的類比電路很實際的 mismatch 水準,算出來剛好就是教科書宣稱的 30 dB 量級——這個「typical」數字不是經驗口訣,而是 small-error 公式的直接結果。: 0.3 dB / 3° is a very realistic level of mismatch for an uncalibrated analogue circuit, and it works out to exactly the 30 dB figure the textbooks quote — that “typical” number is not a rule of thumb but a direct consequence of the small-error formula.:0.3 dB/3° は未校正のアナログ回路として非常に現実的なミスマッチ水準であり、計算するとちょうど教科書が挙げる 30 dB になる——この「典型値」は経験則ではなく、微小誤差の式から直接導かれる帰結である。

(b) RC–CR 漂移 +20% 的後果(b) The consequence of a +20% RC–CR drift(b) RC–CR が +20% ドリフトした場合の帰結

已知/目標Given / required既知/目標:RC–CR 90° 網路的 corner $1/RC$ 因 process 漂移 +20%(即 $RC\to 1.2\,RC$),訊號頻率 $\omega$ 不變(設計時 $\omega=1/RC_{nom}$)。求兩支路的振幅、相位變化與 IRR,並判斷是 amplitude 還是 phase error 主導。: the corner $1/RC$ of an RC–CR 90° network drifts by +20% with process (i.e. $RC\to 1.2\,RC$), while the signal frequency $\omega$ is unchanged (designed with $\omega=1/RC_{nom}$). Find the change in amplitude and phase of the two branches and the resulting IRR, and decide whether amplitude or phase error dominates.:RC–CR 90° 回路網のコーナー $1/RC$ がプロセスばらつきで +20% ドリフトし(すなわち $RC\to 1.2\,RC$)、信号周波数 $\omega$ は変わらない(設計時は $\omega=1/RC_{nom}$)。2 つの枝の振幅と位相の変化、および結果として得られる IRR を求め、振幅誤差と位相誤差のどちらが支配的かを判定せよ。

公式與來源Formulas and sources公式と出典:兩支路的轉移函數:: the transfer functions of the two branches::2 つの枝の伝達関数は次のとおり:

$$H_{RC}=\frac{1}{1+j\omega RC},\qquad H_{CR}=\frac{j\omega RC}{1+j\omega RC},\qquad \frac{H_{CR}}{H_{RC}}=j\omega RC$$

比值 $j\omega RC$ 一眼給出兩個結論:相位差恆為 $\arg(j)=90°$(與頻率無關);振幅比為 $|\omega RC|$(只在 $\omega=1/RC$ 時等於 1)。The ratio $j\omega RC$ gives two conclusions at a glance: the phase difference is always $\arg(j)=90°$ (independent of frequency); and the amplitude ratio is $|\omega RC|$ (equal to 1 only when $\omega=1/RC$).比 $j\omega RC$ を見れば 2 つの結論が一目で分かる:位相差は常に $\arg(j)=90°$ であり(周波数に依存しない)、振幅比は $|\omega RC|$ である($\omega=1/RC$ のときだけ 1 になる)。

逐步代入Substituting step by step順に代入:

  • 漂移後 $\omega RC=\omega\times 1.2\,RC_{nom}=1.2$ → 振幅比 1.2,即 $\varepsilon=0.2$(20% gain error)after the drift $\omega RC=\omega\times 1.2\,RC_{nom}=1.2$ → an amplitude ratio of 1.2, i.e. $\varepsilon=0.2$ (a 20% gain error)ドリフト後は $\omega RC=\omega\times 1.2\,RC_{nom}=1.2$ → 振幅比 1.2、すなわち $\varepsilon=0.2$(20% の利得誤差)
  • 相位差仍是 90° → $\theta=0$,phase error 為零(結構保證)the phase difference is still 90° → $\theta=0$, zero phase error (guaranteed by the structure)位相差は依然として 90° → $\theta=0$ で位相誤差はゼロ(構造的に保証される)
  • $\text{IRR}\approx\dfrac{\varepsilon^2}{4}=\dfrac{0.2^2}{4}=\dfrac{0.04}{4}=0.01$ → $10\log(0.01)=\mathbf{-20\,dB}$
It is the AMPLITUDE imbalance of the drifted RC–CR (not phase) that destroys Hartley's IRR — weakness (2)/(3) listed under HW12.

物理詮釋Physical interpretation物理的な解釈:一個尋常的 ±20% RC process 變異就把 IRR 從 −30 dB 砍到 −20 dB——而且輸不在 90° 相位(那是結構完美的),輸在振幅。這就是 Hartley 必須加 tuning/limiting 或乾脆換架構的原因。: an everyday ±20% RC process variation slashes the IRR from −30 dB to −20 dB — and the loss is not in the 90° phase (which is structurally perfect) but in the amplitude. This is why a Hartley receiver needs tuning or limiting, or a different architecture altogether.:ごく普通の ±20% の RC プロセスばらつきだけで IRR は −30 dB から −20 dB まで落ちる——しかも負けているのは 90° の位相(構造的に完璧)ではなく振幅のほうである。これが Hartley に tuning/limiting を足すか、いっそアーキテクチャを乗り換えねばならない理由である。

(c) Weaver 的 $\omega_2$ 選擇與 secondary image(c) Choosing $\omega_2$ in a Weaver receiver, and the secondary image(c) Weaver における $\omega_2$ の選び方と secondary image

已知/目標Given / required既知/目標:RF $=2.4\,\text{GHz}$、$\omega_1=2.3\,\text{GHz}$(low-side,IF₁ $=100\,\text{MHz}$)、目標 IF₂ $=20\,\text{MHz}$。比較 $\omega_2=120\,\text{MHz}$ 與 $80\,\text{MHz}$ 兩種選法(皆滿足 $|100-\omega_2|=20$):secondary image 在 IF₁ 的哪裡?inter-stage LPF 救得了哪一個?: RF $=2.4\,\text{GHz}$, $\omega_1=2.3\,\text{GHz}$ (low-side, IF₁ $=100\,\text{MHz}$), target IF₂ $=20\,\text{MHz}$. Compare $\omega_2=120\,\text{MHz}$ with $80\,\text{MHz}$ (both satisfy $|100-\omega_2|=20$): where does the secondary image fall in IF₁, and which one can the inter-stage LPF save?:RF $=2.4\,\text{GHz}$、$\omega_1=2.3\,\text{GHz}$(low-side、IF₁ $=100\,\text{MHz}$)、目標 IF₂ $=20\,\text{MHz}$。$\omega_2=120\,\text{MHz}$ と $80\,\text{MHz}$ の 2 つの選び方(いずれも $|100-\omega_2|=20$ を満たす)を比較する:secondary image は IF₁ のどこに落ちるか、inter-stage LPF はどちらを救えるか。

公式與來源Formulas and sources公式と出典:第二次混頻的 image 是 desired IF₁ 對 $\omega_2$ 的鏡像(同 S.1(d) 的原理,套用在 Weaver 的第二級):: the image of the second mix is the mirror of the desired IF₁ about $\omega_2$ (the same principle as S.1(d), applied to Weaver's second stage)::2 回目のミキシングの image は、desired IF₁ を $\omega_2$ について折り返した像である(S.1(d) と同じ原理を Weaver の第 2 段に適用したもの):

$$f_{im,IF1}=2\omega_2-f_{IF1}$$

它對應的 RF 為 $\omega_1+f_{im,IF1}$(low-side 第一混頻)。The corresponding RF is $\omega_1+f_{im,IF1}$ (low-side first mix).対応する RF は $\omega_1+f_{im,IF1}$ となる(low-side の第 1 ミキシング)。

逐步代入Substituting step by step順に代入:

  • $\omega_2=120\,\text{MHz}$:$f_{im,IF1}=2\times 120-100=\mathbf{140\,MHz}$(RF $=2.3+0.14=2.44\,\text{GHz}$)。140 MHz 在 desired 100 MHz $\omega_2=120\,\text{MHz}$: $f_{im,IF1}=2\times 120-100=\mathbf{140\,MHz}$ (RF $=2.3+0.14=2.44\,\text{GHz}$). 140 MHz lies $\omega_2=120\,\text{MHz}$:$f_{im,IF1}=2\times 120-100=\mathbf{140\,MHz}$(RF $=2.3+0.14=2.44\,\text{GHz}$)。140 MHz は desired の 100 MHz 之上aboveより上 → cutoff ≈ 110 MHz 的 inter-stage LPF 通過 100、壓掉 140 ✓ the desired 100 MHz → an inter-stage LPF with a cutoff of ≈ 110 MHz passes 100 and suppresses 140 ✓にある → cutoff ≈ 110 MHz の inter-stage LPF なら 100 を通し 140 を落とせる ✓
  • $\omega_2=80\,\text{MHz}$:$f_{im,IF1}=2\times 80-100=\mathbf{60\,MHz}$(RF $=2.3+0.06=2.36\,\text{GHz}$)。60 MHz 在 desired $\omega_2=80\,\text{MHz}$: $f_{im,IF1}=2\times 80-100=\mathbf{60\,MHz}$ (RF $=2.3+0.06=2.36\,\text{GHz}$). 60 MHz lies $\omega_2=80\,\text{MHz}$:$f_{im,IF1}=2\times 80-100=\mathbf{60\,MHz}$(RF $=2.3+0.06=2.36\,\text{GHz}$)。60 MHz は desired より之下below——任何讓 100 MHz 通過的 LPF 必然也讓 60 MHz 通過。Unfilterable ✗ the desired signal — and any LPF that passes 100 MHz must necessarily pass 60 MHz too. Unfilterable ✗にある——100 MHz を通す LPF は必ず 60 MHz も通してしまう。Unfilterable ✗
Choose ω₂ above IF₁ so the secondary image lands above the desired signal where the LPF can kill it — "must avoid secondary image" (6.2 Weaver figure).

物理詮釋Physical interpretation物理的な解釈:兩種 $\omega_2$ 給出一樣的 IF₂,差別全在 image 的落點——high-side 第二 LO 把 image 推到 LPF 的 stopband,low-side 把它藏進 passband。Weaver 的 frequency plan 自由度必須用來避開 secondary image。: the two choices of $\omega_2$ give the same IF₂; the entire difference is where the image lands — a high-side second LO pushes the image into the LPF's stopband, while a low-side one hides it inside the passband. Weaver's freedom in frequency planning has to be spent on dodging the secondary image.:2 つの $\omega_2$ はどちらも同じ IF₂ を与え、違いは image の落ちる位置だけである——high-side の第 2 LO は image を LPF の stopband へ押し出し、low-side はそれを passband の中に隠してしまう。Weaver の frequency plan の自由度は、secondary image を避けるために使わなければならない。

(d) Weaver vs Hartley 總結(d) Weaver vs Hartley in summary(d) Weaver と Hartley のまとめ

Weaver 勝在哪Where Weaver winsWeaver の強み:它用第二組 quadrature mixer 取代 RC–CR 網路完成 90° 相移——LO 的相位由分頻/多相電路產生、不依賴會漂移的 $RC$ 絕對值((b) 的 −20 dB 災難不存在),且兩路輸出可在 current domain 直接相加。: it uses a second pair of quadrature mixers instead of an RC–CR network to accomplish the 90° phase shift — the LO phase comes from a divider or polyphase circuit and does not depend on the absolute value of a drifting $RC$ (the −20 dB disaster of part (b) simply does not arise), and the two outputs can be summed directly in the current domain.:RC–CR 網の代わりに 2 組目の quadrature mixer で 90° 移相を実現する点——LO の位相は分周器やポリフェーズ回路が生成し、ドリフトする $RC$ の絶対値に依存しない((b) の −20 dB という惨事は起こらない)。しかも 2 系統の出力は current domain でそのまま加算できる。

共同極限Their shared limit共通の限界:兩者都靠 quadrature 精度做 image cancellation,所以: both rely on quadrature accuracy for image cancellation, so :どちらも image cancellation を quadrature 精度に頼るので、都被同一條 $(\varepsilon^2+\theta^2)/4$ 公式綁住both are bound by the very same $(\varepsilon^2+\theta^2)/4$ formula同じ $(\varepsilon^2+\theta^2)/4$ の式に縛られる——mixer 的 I/Q matching(典型 30–40 dB)才是天花板。這正是 low-IF receiver 把訊號數位化後再用 digital image rejection 補完最後 20–30 dB 的原因。— the mixer's I/Q matching (typically 30–40 dB) is the ceiling. That is precisely why a low-IF receiver digitizes the signal and then makes up the last 20–30 dB with digital image rejection.——mixer の I/Q matching(典型的に 30–40 dB)が天井になる。low-IF receiver が信号をディジタル化してから digital image rejection で最後の 20–30 dB を稼ぐのは、まさにこのためである。

S.4 Design
1.96 GHz transmitter — carrier feedthrough, LO pulling, offset-LO, two-step
Synthesis of 6.3: SSB up-mixer, direct-conversion TX, feedthrough & pulling, offset LO, two-step TX.
Spec Direct-conversion TX, carrier 1.96 GHz, PA output +30 dBm
Baseband I/Q amplitude 400 mV; DAC offset on the I path 4 mV; carrier-feedthrough spec ≤ −45 dBc
PA-to-VCO isolation 50 dB

(a) Compute the carrier feedthrough. Does it meet spec? What offset is allowed?

(b) Explain why a VCO running at 1.96 GHz is in trouble here (numbers), and when the pulling is worst.

(c) Design the offset-LO fix with $\omega_2 = \omega_1/2$: give both frequencies, the mixer products to filter, and why neither VCO is pulled.

(d) Give the two-step alternative with IF₁ = 400 MHz: second LO, the image of the second up-mix, the filter that removes it, and the I/Q benefit.

Answer

(a) Carrier feedthrough

已知/目標Given / required既知/目標:baseband I/Q 振幅 $V_{sig}=400\,\text{mV}$、I-path DAC offset $V_{os}=4\,\text{mV}$、spec ≤ −45 dBc。求 feedthrough、判斷是否達標、反推允許的 offset。: baseband I/Q amplitude $V_{sig}=400\,\text{mV}$, I-path DAC offset $V_{os}=4\,\text{mV}$, spec ≤ −45 dBc. Find the feedthrough, decide whether it meets the spec, and work back to the offset allowed.:baseband I/Q 振幅 $V_{sig}=400\,\text{mV}$、I 側 DAC の offset $V_{os}=4\,\text{mV}$、spec は ≤ −45 dBc。feedthrough を求め、spec を満たすか判定し、許容される offset を逆算する。

公式與來源Formulas and sources公式と出典:up-mixer 把 baseband 乘上 LO;baseband 的 DC offset 也被乘上去,直接以 carrier 頻率出現在輸出(carrier feedthrough = DC offset × LO,6.3)。相對載波功率:: the up-mixer multiplies baseband by the LO, and the baseband DC offset gets multiplied too, appearing at the output directly at the carrier frequency (carrier feedthrough = DC offset × LO, 6.3). Relative to the carrier power::up-mixer は baseband に LO を掛け算する。baseband の DC offset も同じく掛け算され、出力にちょうど carrier 周波数で現れる(carrier feedthrough = DC offset × LO、6.3)。carrier 電力に対する相対値は:

$$\text{feedthrough}\;[\text{dBc}]=20\log\frac{V_{os}}{V_{sig}}$$

逐步代入Substituting step by step順に代入:

  • $20\log\dfrac{4}{400}=20\log(0.01)=\mathbf{-40\,dBc}$ — fails the −45 dBc spec by 5 dB
  • 反推允許值:$V_{os}\leq V_{sig}\times 10^{-45/20}=400\times 5.62\times 10^{-3}=2.25\,\text{mV}$working back to the allowed value: $V_{os}\leq V_{sig}\times 10^{-45/20}=400\times 5.62\times 10^{-3}=2.25\,\text{mV}$許容値の逆算:$V_{os}\leq V_{sig}\times 10^{-45/20}=400\times 5.62\times 10^{-3}=2.25\,\text{mV}$
Feedthrough = −40 dBc(差 5 dB 不達標);允許 offset $V_{os}\leq$ ≈ 2.2 mV → trim the DACs / digital offset calibrationFeedthrough = −40 dBc (fails the spec by 5 dB); the allowed offset is $V_{os}\leq$ ≈ 2.2 mV → trim the DACs / digital offset calibrationFeedthrough = −40 dBc(5 dB 足りず spec 未達);許容 offset は $V_{os}\leq$ ≈ 2.2 mV → trim the DACs / digital offset calibration

物理詮釋Physical interpretation物理的な解釈:1% 的 DC offset 就是 −40 dBc——feedthrough 對 offset 是 20 dB/decade 的線性靈敏度,所以 −45 dBc 等於要求 offset 壓到滿幅的 0.56% 以下,類比 trimming 之外通常還要數位校準。這個未調變的 carrier 殘留會打壞接收端的 constellation 原點,等效於 RX 的 DC offset 問題。: a 1% DC offset is already −40 dBc — feedthrough tracks offset with a linear 20 dB/decade sensitivity, so −45 dBc means holding the offset below 0.56% of full scale, which usually needs digital calibration on top of analogue trimming. This residual unmodulated carrier corrupts the origin of the constellation at the receiver, equivalent to the DC-offset problem on the RX side.:1% の DC offset がもう −40 dBc に相当する——feedthrough は offset に対して 20 dB/decade の線形な感度をもつので、−45 dBc とは offset をフルスケールの 0.56% 以下に抑えよという要求であり、アナログの trimming に加えてディジタル校正がふつう必要になる。この無変調の残留 carrier は受信側で constellation の原点を狂わせ、RX 側の DC オフセット問題と等価になる。

(b) Injection pulling 的數字(b) Putting numbers on injection pulling(b) Injection pulling を数字で見る

已知/目標Given / required既知/目標:PA 輸出 $+30\,\text{dBm}$、PA→VCO isolation 50 dB、direct-conversion 表示 VCO 也跑在 1.96 GHz。評估 VCO 的處境與最糟時機。: PA output $+30\,\text{dBm}$, PA→VCO isolation 50 dB, and direct conversion means the VCO also runs at 1.96 GHz. Assess the VCO's predicament and the worst moment for it.:PA 出力 $+30\,\text{dBm}$、PA→VCO の isolation 50 dB、direct-conversion なので VCO も 1.96 GHz で動く。VCO の置かれた状況と最悪のタイミングを評価する。

公式與來源Formulas and sources公式と出典:注入訊號功率 $=P_{PA}-\text{isolation}$;當注入頻率落在 oscillator 的 lock range 內(這裡 PA 頻譜「就是」VCO 頻率),oscillator 會被 injection pulling/locking 拖走(6.3,Adler 型行為:pulling 強度隨注入比 $\sqrt{P_{inj}/P_{osc}}$ 增大)。: the injected signal power is $=P_{PA}-\text{isolation}$; when the injection frequency falls within the oscillator's lock range (here the PA spectrum *is* the VCO frequency), the oscillator is dragged by injection pulling/locking (6.3, Adler-type behaviour: the strength of the pulling grows with the injection ratio $\sqrt{P_{inj}/P_{osc}}$).:注入される信号電力は $=P_{PA}-\text{isolation}$。注入周波数が oscillator の lock range に入ると(ここでは PA のスペクトルが「まさに」VCO 周波数である)、oscillator は injection pulling/locking で引きずられる(6.3、Adler 型の振る舞い:pulling の強さは注入比 $\sqrt{P_{inj}/P_{osc}}$ とともに増大する)。

逐步代入Substituting step by step順に代入:

  • 注入到 VCO 的 PA 能量:$+30-50=-20\,\text{dBm}=10\,\mu\text{W}$PA energy injected into the VCO: $+30-50=-20\,\text{dBm}=10\,\mu\text{W}$VCO に注入される PA のエネルギー:$+30-50=-20\,\text{dBm}=10\,\mu\text{W}$
  • VCO core 振盪功率約 $0\,\text{dBm}=1\,\text{mW}$ → 注入比 $-20\,\text{dB}$(電壓比 $1/10$)the VCO core oscillates at roughly $0\,\text{dBm}=1\,\text{mW}$ → an injection ratio of $-20\,\text{dB}$ (a voltage ratio of $1/10$)VCO コアの発振電力はおよそ $0\,\text{dBm}=1\,\text{mW}$ → 注入比は $-20\,\text{dB}$(電圧比 $1/10$)
  • 對 oscillator 而言 −20 dB 的同頻注入是非常強的擾動(pulling 實驗通常 −40 dB 以下才可忽略)→ VCO 相位被 PA 的調變頻譜拖著跑,modulation 被自己的輸出污染for an oscillator, a co-frequency injection at −20 dB is an extremely strong disturbance (pulling experiments usually consider it negligible only below −40 dB) → the VCO phase is dragged along by the PA's modulation spectrum, and the modulation is contaminated by its own outputoscillator にとって −20 dB の同一周波数注入は極めて強い擾乱である(pulling の実験では −40 dB 以下でようやく無視できる)→ VCO の位相は PA の変調スペクトルに引きずられ、変調が自分自身の出力で汚染される

物理詮釋Physical interpretation物理的な解釈:最糟的時刻是 burst ramp-up(PA 包絡劇烈變動、注入量瞬間拉滿)以及頻譜被調變展寬時——偏偏這正是資料在傳的時候,所以 pulling 直接表現為 EVM/頻譜劣化,無法靠事後校準救回。Shielding 只能買到有限的 isolation,治本要靠 (c)(d) 的 frequency planning。: the worst moment is during burst ramp-up (the PA envelope swings violently and the injection peaks) and when the spectrum is broadened by modulation — which is exactly when data is being transmitted, so pulling shows up directly as EVM and spectral degradation that no amount of after-the-fact calibration can undo. Shielding buys only so much isolation; the real cure is the frequency planning of parts (c) and (d).:最悪なのは burst の ramp-up 時(PA の包絡線が激しく振れ、注入量が一気に跳ね上がる)と、変調でスペクトルが広がっているとき——それはまさにデータを送っている最中なので、pulling は EVM やスペクトル劣化としてそのまま現れ、後からの校正では取り戻せない。シールドで買える isolation には限りがあり、根本的な処方は (c)(d) の frequency planning である。

(c) Offset-LO 設計($\omega_2=\omega_1/2$)(c) An offset-LO design ($\omega_2=\omega_1/2$)(c) Offset-LO の設計($\omega_2=\omega_1/2$)

已知/目標Given / required既知/目標:要求 $\omega_1+\omega_2=\omega_c=1.96\,\text{GHz}$ 且 $\omega_2=\omega_1/2$。求兩個頻率、offset mixer 的輸出產物與濾波、以及為何兩個 VCO 都不被 pull。: require $\omega_1+\omega_2=\omega_c=1.96\,\text{GHz}$ with $\omega_2=\omega_1/2$. Find the two frequencies, the offset mixer's output products and their filtering, and explain why neither VCO gets pulled.:$\omega_1+\omega_2=\omega_c=1.96\,\text{GHz}$ かつ $\omega_2=\omega_1/2$ を要求する。2 つの周波数、offset mixer の出力成分とその濾波、そしてなぜどちらの VCO も pull されないのかを求める。

公式與來源Formulas and sources公式と出典:LO chain 先用 mixer 把 $\omega_1$ 與 $\omega_2$ 相乘產生 $\omega_1\pm\omega_2$,BPF 選出和頻當作真正的 up-conversion LO(6.3 offset-LO 圖);$\omega_2$ 由 $\omega_1$ 經 ÷2 分頻產生,只需一個 VCO。: the LO chain first mixes $\omega_1$ with $\omega_2$ to produce $\omega_1\pm\omega_2$, and a BPF selects the sum as the real up-conversion LO (the 6.3 offset-LO figure); $\omega_2$ is generated from $\omega_1$ by a ÷2 divider, so only one VCO is needed.:LO chain はまず mixer で $\omega_1$ と $\omega_2$ を掛けて $\omega_1\pm\omega_2$ を作り、BPF で和の成分を選んで実際の up-conversion 用 LO とする(6.3 の offset-LO の図)。$\omega_2$ は $\omega_1$ を ÷2 分周して得るので、VCO は 1 個で済む。

逐步代入Substituting step by step順に代入:

  • $\omega_1+\dfrac{\omega_1}{2}=\dfrac{3}{2}\omega_1=1.96\,\text{GHz}$ → $\omega_1=\dfrac{2}{3}\times 1.96=\mathbf{1.307\,GHz}$,$\omega_2=\dfrac{1}{3}\times 1.96=\mathbf{653\,MHz}$$\omega_1+\dfrac{\omega_1}{2}=\dfrac{3}{2}\omega_1=1.96\,\text{GHz}$ → $\omega_1=\dfrac{2}{3}\times 1.96=\mathbf{1.307\,GHz}$, $\omega_2=\dfrac{1}{3}\times 1.96=\mathbf{653\,MHz}$$\omega_1+\dfrac{\omega_1}{2}=\dfrac{3}{2}\omega_1=1.96\,\text{GHz}$ → $\omega_1=\dfrac{2}{3}\times 1.96=\mathbf{1.307\,GHz}$、$\omega_2=\dfrac{1}{3}\times 1.96=\mathbf{653\,MHz}$
  • offset-mixer 輸出:$\omega_1+\omega_2=1.96\,\text{GHz}$(wanted);$\omega_1-\omega_2=1.307-0.653=653\,\text{MHz}$(unwanted,離載波 1.3 GHz → BPF 很好做);另有 harmonic 產物($2\omega_1=2.61\,\text{GHz}$、$2\omega_2=1.307\,\text{GHz}$、…)也都遠離 1.96 GHzoffset-mixer outputs: $\omega_1+\omega_2=1.96\,\text{GHz}$ (wanted); $\omega_1-\omega_2=1.307-0.653=653\,\text{MHz}$ (unwanted, 1.3 GHz from the carrier → an easy BPF); plus harmonic products ($2\omega_1=2.61\,\text{GHz}$, $2\omega_2=1.307\,\text{GHz}$, …), all far from 1.96 GHzoffset-mixer の出力:$\omega_1+\omega_2=1.96\,\text{GHz}$(wanted);$\omega_1-\omega_2=1.307-0.653=653\,\text{MHz}$(unwanted、carrier から 1.3 GHz 離れており BPF は容易);さらに高調波成分($2\omega_1=2.61\,\text{GHz}$、$2\omega_2=1.307\,\text{GHz}$、…)もすべて 1.96 GHz から遠い
Neither 1.307 GHz nor 653 MHz coincides with the PA's 1.96 GHz spectrum → injection pulling is broken by frequency offset, not by shielding.

物理詮釋Physical interpretation物理的な解釈:PA 的大訊號頻譜集中在 1.96 GHz,而系統裡: the PA's large-signal spectrum is concentrated at 1.96 GHz, and in this system :PA の大信号スペクトルは 1.96 GHz に集中しているが、このシステムには沒有任何 oscillator 跑在 1.96 GHzno oscillator runs at 1.96 GHz at all1.96 GHz で動く oscillator が 1 つも存在しない——注入訊號落在兩個 VCO 的 lock range 之外,pulling 從機制上被消滅。1.96 GHz 只存在於 mixer 輸出這種「非再生」節點上,不怕被注入。— the injected signal falls outside the lock range of both VCOs, so pulling is eliminated at the mechanism level. 1.96 GHz exists only at non-regenerative nodes such as the mixer output, which cannot be injection-locked.——注入信号は 2 つの VCO のどちらの lock range からも外れるので、pulling は機構のレベルで消滅する。1.96 GHz は mixer 出力のような「非再生的」なノードにしか存在せず、注入引き込みを受けない。

(d) Two-step TX(IF₁ = 400 MHz)(d) A two-step TX (IF₁ = 400 MHz)(d) Two-step TX(IF₁ = 400 MHz)

已知/目標Given / required既知/目標:先 quadrature up-mix 到 IF₁ $=400\,\text{MHz}$,再上變頻到 1.96 GHz。求第二個 LO、第二次 up-mix 的 image、負責的 filter,與 I/Q 上的好處。: quadrature up-mix first to IF₁ $=400\,\text{MHz}$, then up-convert to 1.96 GHz. Find the second LO, the image of the second up-mix, the filter responsible for it, and the benefit on the I/Q side.:まず quadrature up-mix で IF₁ $=400\,\text{MHz}$ に上げ、次に 1.96 GHz へ上変換する。第 2 の LO、2 回目の up-mix の image、それを取り除くフィルタ、そして I/Q 上の利点を求める。

公式與來源Formulas and sources公式と出典:第二級 mixer 產生 $f_2\pm f_{IF1}$ 兩個 sideband:和頻是 carrier,差頻就是 TX 端的 image,落在離 carrier $2f_{IF1}$ 處——與 RX image problem 同構,靠 mixer 後的 BPF 解決(6.3 two-step 圖)。: the second mixer produces the two sidebands $f_2\pm f_{IF1}$: the sum is the carrier, and the difference is the TX-side image, landing $2f_{IF1}$ from the carrier — isomorphic to the RX image problem, and solved by a BPF after the mixer (the 6.3 two-step figure).:第 2 段の mixer は $f_2\pm f_{IF1}$ の 2 つの sideband を生む:和が carrier、差が TX 側の image で、carrier から $2f_{IF1}$ 離れた位置に落ちる——RX の image problem と同型であり、mixer 後段の BPF で解決する(6.3 の two-step の図)。

逐步代入Substituting step by step順に代入:

  • $f_2=1.96-0.4=\mathbf{1.56\,GHz}$
  • image(差頻):$f_2-f_{IF1}=1.56-0.4=\mathbf{1.16\,GHz}$,在 carrier 下方 $2f_{IF1}=800\,\text{MHz}$ 處image (the difference term): $f_2-f_{IF1}=1.56-0.4=\mathbf{1.16\,GHz}$, sitting $2f_{IF1}=800\,\text{MHz}$ below the carrierimage(差成分):$f_2-f_{IF1}=1.56-0.4=\mathbf{1.16\,GHz}$、carrier の $2f_{IF1}=800\,\text{MHz}$ 下に位置する
  • $800\,\text{MHz}/1.96\,\text{GHz}\approx 41\%$ 的 fractional offset → post-mixer BPF 輕鬆壓制a fractional offset of $800\,\text{MHz}/1.96\,\text{GHz}\approx 41\%$ → easily suppressed by the post-mixer BPF$800\,\text{MHz}/1.96\,\text{GHz}\approx 41\%$ という比率のオフセット → post-mixer BPF で楽に抑圧できる
f₂ = 1.56 GHz;image at 1.16 GHz(2·IF₁ = 800 MHz below carrier)→ removed by the inter-stage / post-mixer BPFf₂ = 1.56 GHz; image at 1.16 GHz (2·IF₁ = 800 MHz below carrier) → removed by the inter-stage / post-mixer BPFf₂ = 1.56 GHz;image は 1.16 GHz(carrier の 2·IF₁ = 800 MHz 下)→ inter-stage / post-mixer BPF で除去

物理詮釋Physical interpretation物理的な解釈:I/Q quadrature 在 400 MHz 建立,匹配遠比 1.96 GHz 容易(mismatch 大致隨頻率上升)→ RF 端的 EVM 與 carrier feedthrough 都更低;同時兩個 oscillator(quadrature LO 在 0.4 GHz、$f_2=1.56\,\text{GHz}$)都不等於 PA 的 1.96 GHz,pulling 同樣被避開。代價是多一顆 LO 與兩個 BPF(6.3)。: the I/Q quadrature is established at 400 MHz, where matching is far easier than at 1.96 GHz (mismatch broadly worsens with frequency) → both EVM and carrier feedthrough at RF are lower; at the same time neither oscillator (the quadrature LO at 0.4 GHz and $f_2=1.56\,\text{GHz}$) sits at the PA's 1.96 GHz, so pulling is avoided as well. The price is one more LO and two more BPFs (6.3).:I/Q の直交は 400 MHz で確立され、1.96 GHz よりはるかに整合を取りやすい(ミスマッチは概ね周波数とともに悪化する)→ RF 側の EVM も carrier feedthrough も低くなる。同時に 2 つの oscillator(0.4 GHz の quadrature LO と $f_2=1.56\,\text{GHz}$)はいずれも PA の 1.96 GHz と一致しないので、pulling も同様に回避できる。代償は LO が 1 個と BPF が 2 個増えることである(6.3)。