Spectral Efficiency of Mixed-ADC Massive MIMO

We study the spectral efficiency (SE) of a mixed-analog-to-digital converter (ADC) massive multiple-input multiple-output (MIMO) system in which <inline-formula><tex-math notation="LaTeX">$K$</tex-math></inline-formula> single-antenna users communicate with a base station equipped with <inline-formula><tex-math notation="LaTeX">$M$ </tex-math></inline-formula> antennas connected to <inline-formula><tex-math notation="LaTeX">$N$</tex-math> </inline-formula> high-resolution ADCs and <inline-formula><tex-math notation="LaTeX">$M-N$</tex-math></inline-formula> 1-b ADCs. This architecture has been proposed as an approach for realizing massive MIMO systems with reasonable power consumption. First, we investigate the effectiveness of mixed-ADC architectures in overcoming the channel estimation error caused by coarse quantization. For the channel estimation phase, we study to what extent one can combat the SE loss by exploiting just <inline-formula><tex-math notation="LaTeX">$N\ll M$</tex-math></inline-formula> pairs of high-resolution ADCs. We extend the round-robin training scheme for mixed-ADC systems to include both high-resolution and 1-b quantized observations. Then, we analyze the impact of the resulting channel estimation error in the data detection phase. We consider random high-resolution ADC assignment and also analyze a simple antenna selection scheme to increase the SE. Analytical expressions are derived for the SE for maximum ratio combining and numerical results are presented for zero-forcing detection. Performance comparisons are made against systems with uniform ADC resolution and against mixed-ADC systems without round-robin training to illustrate under what conditions each approach provides the greatest benefit.