A Three-Stage-Concatenated Non-Linear MMSE Interference Rejection Combining Aided MIMO-OFDM Receiver and its EXIT-Chart Analysis

The demodulation reference signal of the 5G Multiple-Input Multiple-Output Orthogonal Frequency-Division Multiplexing (MIMO-OFDM) waveform has been designed for supporting Minimum Mean-Square Error-Interference Rejection Combining (MMSE-IRC) equalization, which has become the state-of-the-art, owing to its enhanced performance in the case of dense frequency reuse, which is typical in 5G. By contrast, in the 4G LTE system, typically turbo equalization techniques were used. The family of Non-Linear receiver techniques tend to be eminently suitable for tough rank-deficient scenarios, when the received signal constellation becomes linearly non-separable. Hence, we propose a novel receiver for interference-constrained MIMO-OFDM systems, relying on a linear MMSE-IRC detector intrinsically amalgamated with an additional NL equalizer. In this way, we may achieve the best of both worlds, retaining the interference rejection capability of the MMSE-IRC detector and the superior performance of the NL equalizer. Our solution circumvents the potential failure of the MMSE-IRC, when the MIMO channels' degree freedom is completely exhausted by the desired users in case the transmitter has a high number of transmission layers for example. Based on this concept, we then design a novel NL equalizer relying on the Smart Ordering and Candidate Adding (SOCA) algorithm. This reduced complexity NL detection algorithm is particularly well suited for practical hardware implementation using parallel processing at a low latency. Briefly, the proposed scheme employs the MMSE-IRC detector for mitigating the interference. It makes the first estimate of the desired user signals and then uses the SOCA detector for further decontaminating the received signals. It also generates the soft information, enabling turbo equalization, wherein iterative detector and decoder iteratively exchange their soft information. We present BLock Error Rate (BLER) results, which show that the proposed scheme can always achieve superior performance to the conventional MMSE-IRC detector at the cost of increasing the complexity. In some cases, our proposed scheme can obtain about 1.5 dB gain, at the cost of 4 times higher complexity. We demonstrate that the complexity of the SOCA detector can be reduced by adjusting its parameterization or at the cost of reducing the self-consistency of the soft information produced by the SOCA detector, which slightly erodes the BLER performance. In order to mitigate this, we propose to use Deep Learning (DL) for enhancing the accuracy of the soft information. Using this technique, we show that the MMSE-IRC-NL-SOCA detector relying on DL attains about 3 dB gain at the cost of only marginally increasing the complexity, compared to the proposed MMSE-IRC-NL-SOCA scheme.