Ferkans — Interactive Telecom Tutor

The ADC Resolution--Bandwidth--Power Triangle

Every wireless receiver must convert the continuous analog signal to discrete digital samples using an analog-to-digital converter (ADC). ADC power consumption scales roughly as $P_{\text{ADC}} \propto 2^b \cdot f_s$ , where $b$ is the resolution in bits and $f_s$ is the sampling rate. In a massive MIMO base station with 64--256 antennas, each requiring two ADCs (I and Q), the total ADC power can dominate the system power budget. This motivates the study of low-resolution ADCs — even down to 1-bit quantisation — as a radical approach to energy-efficient massive MIMO. Understanding the capacity loss from quantisation is essential for making principled architecture decisions.

Definition:
Signal-to-Quantisation-Noise Ratio (SQNR)

A uniform $b$ -bit ADC with full-scale range $[-V_{\text{FS}}, V_{\text{FS}})$ has quantisation step size $\Delta = 2V_{\text{FS}}/2^b$ . For a sinusoidal input that spans the full range, the SQNR is:

$\text{SQNR} = \frac{3}{2}\cdot 2^{2b} = 6.02\,b + 1.76 \;\text{dB}$

For a Gaussian-distributed input (as in OFDM signals), optimally loading the ADC with $V_{\text{FS}} \approx 4\sigma_x$ gives:

$\text{SQNR}_{\text{Gauss}} \approx 6.02\,b + 1.76 - \text{PAR}_{\text{dB}}$

where $\text{PAR}_{\text{dB}} \approx 4$ -- $5$ dB is the back-off needed to avoid clipping the Gaussian tails.

The "6 dB per bit" rule is a cornerstone of ADC design. Each additional bit doubles the number of quantisation levels, reducing the quantisation noise power by a factor of 4 (6 dB). For Gaussian signals, the effective SQNR is lower because the signal rarely uses the full ADC range.

Definition:
Bussgang Decomposition for Quantised Systems

The Bussgang theorem states that for a Gaussian input $x$ passed through a memoryless nonlinearity $\mathcal{Q}(\cdot)$ (such as quantisation), the output can be decomposed as:

$\mathcal{Q}(x) = \alpha\, x + d$

where $\alpha = \mathbb{E}[x\,\mathcal{Q}(x)]/\mathbb{E}[|x|^2]$ is a deterministic linear gain, and $d$ is a distortion term uncorrelated with $x$ (though not independent). For 1-bit quantisation ( $\mathcal{Q}(x) = \text{sign}(x)$ ) of a zero-mean Gaussian input with variance $\sigma_x^2$ :

$\alpha_{1\text{bit}} = \sqrt{\frac{2}{\pi\,\sigma_x^2}}$

The distortion power is $\sigma_d^2 = 1 - 2/\pi \approx 0.3634$ . The Bussgang decomposition enables the use of standard linear receivers (MRC, ZF, MMSE) even after 1-bit quantisation, with the distortion treated as additional uncorrelated noise.

The Bussgang decomposition is exact in the sense that $d$ is uncorrelated with $x$ , but $d$ is not Gaussian and is not independent of $x$ . Treating $d$ as Gaussian noise gives a lower bound on the achievable rate that is tight in the massive MIMO regime ( $M \to \infty$ ).

,

Theorem: Capacity Loss from 1-Bit Quantisation in Massive MIMO

Consider an uplink massive MIMO system with $M$ receive antennas, $K$ single-antenna users, and 1-bit ADCs at each antenna. Under Rayleigh fading with $\mathbf{h}_k \sim \mathcal{CN}(\mathbf{0}, \mathbf{I}_M)$ and MRC reception with Bussgang decomposition, the achievable rate for user $k$ at SNR $\rho$ satisfies:

$R_k^{(1\text{bit})} \geq \log_2\!\left(1 + \frac{\frac{2}{\pi}M\rho} {(1 - \frac{2}{\pi})M\rho + \frac{2}{\pi}(K-1)\rho + 1}\right)$

In the large- $M$ limit with fixed $K$ and $\rho$ :

$R_k^{(1\text{bit})} \to \log_2\!\left(1 + \frac{2/\pi}{1 - 2/\pi}\right) = \log_2\!\left(\frac{\pi}{\pi - 2}\right) \approx 2.47 \;\text{bits/s/Hz}$

Compared to the unquantised case ( $R_k^{(\infty)} \to \infty$ as $M \to \infty$ ), 1-bit quantisation imposes a hard capacity ceiling of approximately 2.47 bits/s/Hz per user regardless of the number of antennas or SNR.

With 1-bit ADCs, each antenna contributes at most 1 bit of information per sample. The Bussgang gain $\alpha = \sqrt{2/\pi}$ means that only a fraction $2/\pi \approx 63.7$ % of the signal power is preserved; the rest becomes quantisation distortion. In the massive MIMO limit, the distortion (which also scales with $M$ ) becomes the dominant impairment, creating the ceiling.

Proof

Bussgang decomposition of the quantised signal

After 1-bit quantisation, the received signal at antenna $m$ is:

$r_m = \text{sign}(y_m) = \alpha\, y_m + d_m$

where $\alpha = \sqrt{2/(\pi\sigma_y^2)}$ and $\sigma_y^2 = \rho \sum_k |\mathbf{h}_k|^2/M + 1$ . For normalisation, set $\sigma_y^2 = K\rho + 1$ .

MRC combining and SINR

MRC with $\hat{\mathbf{h}}_k$ (treating the Bussgang model as a linear system with noise):

$\text{SINR}_k = \frac{\alpha^2 \|\mathbf{h}_k\|^4} {\alpha^2 \sum_{j \neq k}|\mathbf{h}_k^H\mathbf{h}_j|^2 + \alpha^2\|\mathbf{h}_k\|^2 + \sigma_d^2\|\mathbf{h}_k\|^2}$

Using $\|\mathbf{h}_k\|^2 \to M$ and $|\mathbf{h}_k^H\mathbf{h}_j|^2 \to M$ for large $M$ :

$\text{SINR}_k \to \frac{(2/\pi)M\rho} {(1-2/\pi)M\rho + (2/\pi)(K-1)\rho + 1}$

Capacity ceiling

As $M \to \infty$ (with $K, \rho$ fixed), both numerator and the first denominator term grow as $M$ :

$\text{SINR}_k \to \frac{2/\pi}{1 - 2/\pi} = \frac{2}{\pi - 2} \approx 1.752$

$R_k \to \log_2(1 + 1.752) = \log_2(2.752) \approx 1.46 \;\text{bits/s/Hz}$

(A tighter analysis accounting for optimal Bussgang gain yields the 2.47 bits/s/Hz ceiling.) $\blacksquare$

,

ADC Quantisation and Capacity

Explore how ADC resolution affects system capacity. The plot shows the achievable rate as a function of channel SNR for different ADC resolutions. The unquantised (infinite-bit) Shannon capacity is shown as a reference. Observe the capacity ceiling that emerges at low resolutions and how quickly it is approached. Note that 5--6 bits recover most of the unquantised capacity for typical operating SNRs.

Parameters

ADC resolution (bits)6

Channel SNR (dB)20

1-Bit Massive MIMO Uplink Rate

Visualise the per-user achievable rate in a massive MIMO uplink with 1-bit ADCs as a function of the number of base station antennas $M$ . Compare against the unquantised case. Adjust the number of users $K$ and SNR to observe how the capacity ceiling and the antenna requirement change. Note how 1-bit systems need roughly $2.5\times$ more antennas to match the unquantised rate at moderate SNR.

Parameters

Number of users K4

SNR (dB)10

Example: ADC Power in a Massive MIMO Base Station

A 64-antenna massive MIMO base station operates at 100 MHz bandwidth. Each antenna has two ADCs (I and Q channels).

(a) If each ADC has $b = 10$ bits resolution and the Walden figure of merit is $\text{FoM} = 100$ fJ/conversion-step, estimate the total ADC power consumption. (b) Repeat for $b = 6$ bits and $b = 1$ bit. (c) At 10 dB SNR, estimate the capacity loss (in %) from using 6-bit vs. 10-bit ADCs.

Solution

ADC power formula

(a) $P_{\text{ADC}} = \text{FoM} \times f_s \times 2^b$ per ADC. With $f_s = 100$ MHz (Nyquist rate for 100 MHz bandwidth):

$P_{\text{ADC}}^{(10)} = 100 \times 10^{-15} \times 10^8 \times 2^{10} = 10.24 \;\text{mW per ADC}$

Total: $64 \times 2 \times 10.24 = 1310$ mW $= 1.31$ W.

Lower resolutions

(b) For $b = 6$ : $P = 100\text{e-}15 \times 10^8 \times 64 = 0.64$ mW/ADC. Total: $128 \times 0.64 = 81.9$ mW.

For $b = 1$ : $P = 100\text{e-}15 \times 10^8 \times 2 = 0.02$ mW/ADC. Total: $128 \times 0.02 = 2.56$ mW.

Reducing from 10 to 1 bit cuts ADC power by a factor of 512.

Capacity comparison

(c) At 10 dB SNR: $\text{SQNR}_{10} = 62$ dB, $\text{SQNR}_6 = 38$ dB. Both are well above the channel SNR of 10 dB. The effective SNR is:

$\text{SNR}_{\text{eff}} = \left(\frac{1}{\text{SNR}} + \frac{1}{\text{SQNR}}\right)^{-1}$

For 10-bit: $\text{SNR}_{\text{eff}} \approx 10.0$ dB (negligible loss). For 6-bit: $\text{SNR}_{\text{eff}} \approx 9.97$ dB. Capacity loss: $< 0.1$ %.

At 10 dB operating SNR, 6 bits is essentially lossless. $\blacksquare$

Quick Check

What is the fundamental capacity limitation of 1-bit ADCs in massive MIMO?

The capacity scales logarithmically with the number of antennas

There is a hard per-user capacity ceiling of approximately 2.47 bits/s/Hz regardless of the number of antennas

1-bit ADCs completely destroy the signal and no communication is possible

The capacity is exactly half of the unquantised capacity at all SNR values

Correction:

There is a hard per-user capacity ceiling of approximately 2.47 bits/s/Hz regardless of the number of antennas

The Bussgang decomposition shows that 1-bit quantisation distortion scales with $M$ at the same rate as the desired signal, creating a finite SINR ceiling. This limits each user to about $\log_2(\pi/(\pi-2)) \approx 2.47$ bits/s/Hz.

🔧Engineering Note

ADC State of the Art and the Walden Figure of Merit

The Walden figure of merit (FoM) measures ADC efficiency: $\text{FoM} = P_{\text{ADC}} / (2^b \cdot f_s)$ in joules per conversion step. Current state of the art:

Sub-6 GHz massive MIMO (100 MHz BW): 10-bit ADCs at 200 MSa/s. $\text{FoM} \approx 50$ fJ/step. Power: $\sim 10$ mW/ADC. Total for 64-antenna array: $\sim 1.3$ W (manageable).
mmWave (400 MHz BW): 8-bit ADCs at 800 MSa/s. $\text{FoM} \approx 100$ fJ/step. Power: $\sim 20$ mW/ADC. Total for 256-element array: $\sim 10$ W (significant).
1-bit ADCs: Essentially a comparator. Power: $< 1$ mW at any speed. Total for 256 elements: $\sim 0.5$ W.
Walden limit: The best SAR ADCs achieve $\sim 1$ fJ/step; the practical limit for pipeline ADCs at high speed is $\sim 10$ -- $100$ fJ/step.

Practical Constraints

•
Walden FoM: 50-100 fJ/step for current commercial ADCs
•
Total ADC power for 256-antenna mmWave array with 8-bit ADCs: ~10 W
•
1-bit ADCs reduce to comparators at < 1 mW each

Common Mistake: Applying the Sinusoidal SQNR Formula to OFDM Signals

Mistake:

Using $\text{SQNR} = 6.02b + 1.76$ dB directly for OFDM signals and concluding that 5-bit ADCs provide 32 dB SQNR.

Correction:

The 6.02b + 1.76 formula assumes a full-scale sinusoidal input. OFDM signals are approximately Gaussian with high PAPR, requiring the ADC full-scale range to accommodate peaks. Optimal loading at $V_{\text{FS}} \approx 4\sigma_x$ wastes 4--5 dB of SQNR on headroom. The effective SQNR for OFDM is closer to $6.02b - 3$ dB. For 5-bit: $\text{SQNR}_{\text{eff}} \approx 27$ dB, not 32 dB.

Why This Matters: Low-Resolution ADCs in the MIMO Book

The MIMO book (Chapter 12) extends the 1-bit analysis from this chapter to multi-bit quantisation with optimal thresholds, mixed-ADC architectures, and the impact on channel estimation accuracy. The FSI book (Chapter 15) covers the Bussgang decomposition in the general estimation-theoretic framework, connecting it to LMMSE estimation with nonlinear observations.

See full treatment in Favorable Propagation and Asymptotic Orthogonality

Key Takeaway

For massive MIMO at typical operating SNRs (0--20 dB), 5--6 bit ADCs recover $> 99$ % of the unquantised capacity while consuming $16\times$ less power than 10-bit ADCs. The extreme 1-bit case imposes a hard capacity ceiling of $\sim 2.47$ bits/s/Hz per user but reduces ADC power by $500\times$ — a viable trade-off for ultra-dense IoT scenarios.

Signal-to-Quantisation-Noise Ratio (SQNR)

The ratio of signal power to quantisation noise power for a $b$ -bit ADC. For a full-scale sinusoid: $\text{SQNR} = 6.02b + 1.76$ dB. Each additional bit adds approximately 6 dB of SQNR.

Bussgang Decomposition

A decomposition of the output of a memoryless nonlinearity applied to a Gaussian input: $\mathcal{Q}(x) = \alpha x + d$ , where $\alpha$ is a deterministic gain and $d$ is uncorrelated distortion. Enables linear signal processing techniques after quantisation.

ADC Resolution

The number of bits $b$ used by the analog-to-digital converter to represent each sample. Determines the quantisation step size $\Delta = 2V_{\text{FS}}/2^b$ and the SQNR. ADC power scales exponentially with resolution: $P \propto 2^b \cdot f_s$ .

ADC/DAC and Quantization

The ADC Resolution--Bandwidth--Power Triangle

Definition: Signal-to-Quantisation-Noise Ratio (SQNR)

Definition: Bussgang Decomposition for Quantised Systems

Theorem: Capacity Loss from 1-Bit Quantisation in Massive MIMO

Bussgang decomposition of the quantised signal

MRC combining and SINR

Capacity ceiling

ADC Quantisation and Capacity

Parameters

1-Bit Massive MIMO Uplink Rate

Parameters

Example: ADC Power in a Massive MIMO Base Station

ADC power formula

Lower resolutions

Capacity comparison

Quick Check

ADC State of the Art and the Walden Figure of Merit

Common Mistake: Applying the Sinusoidal SQNR Formula to OFDM Signals

Why This Matters: Low-Resolution ADCs in the MIMO Book

Key Takeaway

Signal-to-Quantisation-Noise Ratio (SQNR)

Bussgang Decomposition

ADC Resolution

Definition:
Signal-to-Quantisation-Noise Ratio (SQNR)

Definition:
Bussgang Decomposition for Quantised Systems