Introduction
In multi-user multiple-input single-output (MU-MISO) systems, Vector Perturbation (VP) is a sophisticated non-linear precoding technique. By adding a complex integer vector to data symbols, VP reduces required transmit power, pushing system performance toward the theoretical capacity of the broadcast channel. However, standard VP schemes are highly sensitive to real-world imperfections, specifically noisy channel state information (CSI) at the transmitter and quantized power scaling factors (PSF) at the receiver.
This research introduces a robust VP precoding scheme that addresses these hurdles by jointly optimizing for both CSI and PSF errors using a Minimum Mean-Square Error (MMSE) criterion. This approach significantly lowers the "error floor" that typically limits high-SNR wireless communications.
System Model and the "Perfectionist" Problem
Most existing research assumes a "perfect" environment: the transmitter knows the exact channel (CSI), and the receiver knows the precise power scaling factor ($\beta$) used for signal normalization. In practice, feedback links are bandwidth-limited, leading to quantized and delayed CSI. Similarly, scaling factors are transmitted over low-rate feed-forward links.
As shown in the system model, the transmitter relies on a feedback link for $\hat{H}$, while the receiver receives an estimated scaling factor $\hat{\beta}$ via a feed-forward link. Even minor inaccuracies in these values cause the system to hit an error floor, where increasing signal power fails to improve the bit error rate (BER) because interference from these inaccuracies dominates the noise.
Robust Joint MMSE Precoding
The authors propose a unified framework to handle these errors simultaneously. The core of their method involves:
- Error Modeling: CSI quantization noise is modeled using rate-distortion theory, while a relative error model is applied to the power scaling factor.
- Joint Optimization: Instead of simply minimizing transmit power, the authors derive a new Mean Square Error (MSE) expression that incorporates the statistical properties of both CSI and PSF noise.
- Closed-Form Solution: They provide optimal expressions for the precoding matrix ($P_r$) and the perturbation vector ($u'_r$).
Figure 5 illustrates the sensitivity of the constellation. When the receiver uses an inaccurate scaling factor ($\hat{\beta}$ instead of $\beta$), the received signal is scaled incorrectly, potentially pushing the symbol outside the decision boundary after the modulo operation. The proposed robust precoder "pre-compensates" for these inevitable receiver-side errors.
Key Insights and Results
A significant finding is that VP precoding is more sensitive to the accuracy of the power scaling factor than to the CSI itself. Improving PSF accuracy through more feedback bits yields non-linear performance gains, whereas CSI improvements generally offer linear gains. Treating these as a joint optimization problem allows the precoder to account for the total error budget of the link.
Numerical simulations using the 3GPP "urban-macro" spatial channel model demonstrate the following:
- Lower Error Floor: The robust precoder maintains a significantly lower BER in high-SNR regions where conventional VP (CVP) fails.
- Superior Efficiency: The proposed scheme can achieve the same BER as conventional methods while requiring approximately 4dB less signal power in specific scenarios.
- Outperformance: The method exceeds the performance of state-of-the-art MMSE-VP schemes that only account for a single type of error.
Conclusion and Future Directions
While the proposed joint optimization offers substantial gains, challenges remain. The search for the optimal perturbation vector relies on sphere decoding, which has exponential complexity relative to the number of users, potentially limiting its use in massive MIMO. Future work may explore adaptive systems that adjust to fluctuating link quality in real-time or investigate the impact of hardware non-linearities, such as power amplifier distortion, in conjunction with these quantization errors.
Summary updated