Two Models for Noisy Feedback in MIMO Channels

Two distinct models of feedback, suited for FDD (Frequency Division Duplex) and TDD (Frequency Division Duplex) systems respectively, have been widely studied in the literature. In this paper, we compare these two models of feedback in terms of the diversity multiplexing tradeoff for varying amount of channel state information at the terminals. We find that, when all imperfections are accounted for, the maximum achievable diversity order in FDD systems matches the diversity order in TDD systems. TDD systems achieve better diversity order at higher multiplexing gains. In FDD systems, the maximum diversity order can be achieved with just a single bit of feedback. Additional bits of feedback (perfect or imperfect) do not affect the diversity order if the receiver does not know the channel state information.

💡 Research Summary

The paper investigates two fundamentally different feedback mechanisms for single‑user MIMO channels—quantized feedback (appropriate for Frequency Division Duplex, FDD) and two‑way training (appropriate for Time Division Duplex, TDD)—and evaluates their impact on the diversity‑multiplexing tradeoff (DMT) under realistic imperfections such as noisy feedback and imperfect channel state information (CSI) at both transmitter and receiver.

System Model
A MIMO link with m transmit and n receive antennas is considered. The forward channel matrix H has i.i.d. complex Gaussian entries and remains constant over a fading block of L channel uses. The transmitter is power‑limited (average SNR). A reciprocal feedback link exists with the same antenna configuration (the receiver becomes the feedback transmitter with n antennas, the original transmitter becomes the feedback receiver with m antennas). For TDD, forward and reverse channels are identical (H = H_f†); for FDD they are independent. Training overhead is ignored in the analysis, assuming L is large enough to accommodate the required training slots (≥ m + n).

DMT Framework
The DMT is expressed via the function G(r, p), which gives the diversity order for a coherent system with multiplexing gain r and power exponent p. G(r, p) = p·G(r·p, 1) and is piecewise linear, connecting points (k·p, p·(m − k)(n − k)) for k = 0,…,min(m,n).

Key Results

1. Perfect CSIR, No Feedback (CSIR) – The classic coherent DMT d_CSIR = G(r, 1) is recovered (Theorem 1).

2. No CSI at Either End (CSI_bR) – Even when both nodes lack CSI and the transmitter first trains the receiver, the DMT remains d = G(r, 1) (Theorem 2). Hence, the presence of CSI at the receiver does not affect the diversity order when the transmitter can train.

3. Quantized Feedback (FDD Model)

   • Perfect CSIR, Perfect Quantized CSIT (CSI_RT_q) – With K quantization levels (K ≥ 1), the diversity is given recursively by B*{m,n,K}(r) = G(r, 1 + B*{m,n,K‑1}(r)), B*_{m,n,0}(r)=0 (Theorem 3).
   • Estimated CSIR, Perfect Quantized CSIT (CSI_bRT_q) – After a training phase, the DMT becomes d = G(r, 1 + G(r, 1)) (Theorem 4). For r → 0 this yields a maximum diversity of mn(mn + 1), i.e., (mn)² larger than the no‑feedback case.
   • Perfect CSIR, Noisy Quantized CSIT (CSIR_bT_q) – With noisy feedback but power‑controlled MAP decoding, the diversity is d = min{B*{m,n,K}(r), max{j≤1+K} B*_{m,n,j}(r)} (Theorem 5). For K = 2 (one bit) the result collapses to d = G(r, 1 + G(r, 1)), identical to the perfect‑feedback case (Corollary 2). As K → ∞ and r → 0 the diversity approaches mn(mn + 2) (Corollary 3).
   • Estimated CSIR, Noisy Quantized CSIT (CSI_bR_bT_q) – The same DMT as the previous case is obtained (Theorem 6, Corollary 4).

   The essential insight is that a single feedback bit suffices to achieve the maximal diversity when the receiver lacks CSI; additional bits do not improve the DMT.

4. Two‑Way Training (TDD Model)

   • If either node possesses perfect CSI, it can train the other perfectly because forward and reverse channels are reciprocal, leading to unbounded diversity (Theorem 7).
   • When both nodes only have imperfect CSI obtained via training, the maximum diversity at zero multiplexing is again mn(mn + 1), identical to the one‑bit quantized‑feedback FDD case. However, for any positive multiplexing gain, the TDD model yields a higher diversity curve than the quantized‑feedback model (see numerical results).

Comparative Conclusions

• Maximum Diversity Equality: When all imperfections (noisy feedback, imperfect CSI) are accounted for, the highest achievable diversity order in FDD (quantized feedback) equals that in TDD (two‑way training).
• Superiority at Higher Multiplexing: TDD outperforms FDD for non‑zero multiplexing gains because reciprocal training can exploit the same channel estimate for both directions, whereas FDD must rely on a limited number of feedback bits.
• Feedback Bit Efficiency: In FDD, a single bit of quantized power‑level feedback is sufficient to reach the diversity ceiling; extra bits are unnecessary if the receiver does not already know the channel.
• Robustness to Feedback Noise: Power‑controlled feedback combined with MAP decoding mitigates the effect of feedback noise, preserving the DMT achieved with perfect feedback.

Practical Implications

• System designers can drastically reduce feedback overhead in FDD systems by employing a one‑bit power‑control scheme, while still attaining the same diversity as more elaborate quantization.
• For applications requiring high spectral efficiency (high multiplexing gain), TDD operation with reciprocal training is preferable, provided hardware can support accurate channel reciprocity.
• The analysis assumes negligible training overhead; in realistic deployments, the tradeoff between training duration and DMT must be evaluated, especially for short coherence times.

Future Directions

• Extending the analysis to multi‑user MIMO (MAC/BC) scenarios where feedback resources are shared among users.
• Investigating the impact of asymmetric power constraints on the feedback link, which is common in uplink‑downlink asymmetry.
• Incorporating finite‑blocklength effects and explicit training overhead into the DMT framework to obtain tighter performance predictions for practical systems.

Overall, the paper provides a rigorous DMT comparison of noisy feedback mechanisms in MIMO, clarifying when and how limited feedback can be sufficient and highlighting the inherent advantage of reciprocal training in TDD systems.