Chapter Summary

Chapter Summary

Key Points

• 1.

  Massive-MIMO receivers are ADC-power-limited at high antenna counts: a Walden-type converter dissipates $P_{\text{ADC}}(b) = c_0\,2^b f_s$, and at $b = 12$, $f_s = 1$ GS/s, $N_r = 256$ the front-end alone exceeds small-cell power budgets. Every bit saved halves ADC power, driving a strong economic case for low-resolution reception.

• 2.

  The 1-bit (sign) quantizer $Q_1(y) = \operatorname{sign}(\operatorname{Re}(y)) + j\,\operatorname{sign}(\operatorname{Im}(y))$ is the extreme case. Its SISO capacity at low SNR loses exactly a factor $2/\pi \approx 0.637$, or about $1.96$ dB, compared with infinite precision (Singh-Mondal-Mehanna-Madhow 2009). At high SNR the loss grows because the 1-bit rate is capped at 1 bit/channel use.

• 3.

  The Bussgang decomposition linearizes any memoryless nonlinearity as $\mathbf{y}_q = \mathbf{B}\mathbf{y} + \mathbf{d}$ with $\mathbf{d} \perp \mathbf{y}$. For the 1-bit quantizer the Bussgang gain is $B_{nn} = \sqrt{2/(\pi[\boldsymbol{\Sigma}_{y}]_{nn})}$ and the quantized covariance obeys the arcsine law $(2/\pi)\arcsin(\rho)$. The residual $\mathbf{d}$ is uncorrelated with the input but is not independent — treating it as Gaussian noise yields a lower bound on mutual information that is tight at low SNR and loose at high SNR.

• 4.

  With Bussgang linearization, the achievable rate of linear combining becomes the standard MMSE expression with two edits: desired and interfering powers scale by $\kappa_b = 1 - \rho_b$, and an extra distortion covariance $\boldsymbol{\Sigma}_{d}$ adds to the noise. Values of $\kappa_b$ follow the Lloyd-Max Gaussian-optimized quantizer: $\kappa_1 = 2/\pi \approx 0.637$, $\kappa_2 \approx 0.882$, $\kappa_4 \approx 0.990$. One additional bit cuts distortion by $\approx 4\times$ — the 6 dB/bit rule.

• 5.

  Mixed-ADC receivers equip a fraction $\alpha$ of antennas with full-resolution ADCs and $(1-\alpha)$ with 1-bit converters. Under channel hardening and MRC the effective array gain becomes $G(\alpha) = \kappa_1 + (1-\kappa_1)\alpha$, linearly interpolating between $\kappa_1$ and $1$. Practical deployments choose $\alpha \in [0.1, 0.3]$, recovering most of the rate at a fraction of the ADC power.

• 6.

  Optimal bit allocation across antennas generalizes the mixed-ADC idea: each antenna $n$ receives its own resolution $b_n$, and the allocation maximizes sum rate subject to $\sum_n 2^{b_n} \leq B_{\max}$. Under the continuous relaxation the optimum is water-filling on log gains, $b_n^\star \propto \tfrac{1}{2}\log_2(g_n\,\text{SNR})$, with the weakest antennas switched off and the strongest getting the majority of the bit budget. The $\tfrac{1}{2}$ slope is the Bussgang analogue of the 6 dB/bit SQNR heuristic.

• 7.

  Constellation design must match the receiver: PSK is the natural input for 1-bit ADC reception because the quantizer discards amplitude information, while QAM's amplitude levels collapse onto the same sign patterns. Massive arrays compensate for the lack of amplitude resolution through coherent combining.

• 8.

  Energy efficiency $\eta = R/P_{\text{rx}}$ is the right deployment metric. Under the ADC-dominated power model the optimal resolution is $b^\star = 1$ at low per-antenna SNR (mmWave uplink), $b^\star \in \{3, 4\}$ at moderate SNR (sub-6 GHz macro), and $b^\star \geq 5$ only at high SNR where conventional receivers dominate. The break-even between 1-bit and 4-bit depends crucially on whether the RF front-end, baseband, or ADC term dominates the denominator — at mmWave the ADC dominates and 1-bit is often the runaway winner.

• 9.

  The CommIT contribution of Jacobsson, Durisi, Coldrey, Studer and Caire (IEEE TWC 2017) is the comprehensive analysis that underpins the rate formulas and mixed-ADC sizing rules used in this chapter. Together with earlier work by Mezghani-Nossek and Mo-Heath, it established low-resolution massive MIMO as a practical alternative to the high-resolution baseline for mmWave and sub-THz systems.