CNN Aided Detection Algorithm for Uplink Multiuser Modulation Encrypted Massive MIMO Systems

In the traditional physical layer encryption methods, both the delay and errors brought by the keys generation and interaction and those from the channel estimation are too high to be employed for the uplink multiuser massive multiple-input and multiple-output (MIMO) systems. Differently, this paper constructs a lightweight encryption scheme with the modulation random chaotic encryption (MRCE) signals in the uplink massive MIMO systems where the keys are not required to be known, a priori, at the base station (BS). Their generation is off-line and does not employ the channel state information, and there is no immediate interaction. Specially, as a deep learning based solution for detecting the MRCE signals in the uplink massive MIMO systems, the convolutional-neural-network aided nonlinear detection (CAD) algorithms are proposed in this paper. The simulation results showed their effectiveness. The anti-eavesdrop ability of the MRCE signals is verified even against eavesdroppers equipped with multiple antennas at high average signal to noise ratio (SNR). As shown in the simulation results, when the BS does not previously know the keys, the proposed CAD algorithms have a much better bit error rate (BER) performance and higher secrecy spectral efficiency (SSE) than the less efficient ones of their corresponding unassisted methods. The BER is closer to the theoretical lower bound of the optimum value obtained by the maximum likelihood method. They require lower average received SNR to converge to the theoretical maximum SSE given by the no error transmissions from the legitimate users to the BS. These performances also approach those of their corresponding unassisted algorithms without prior known keys. The proposed CAD algorithms have medium/strong robustness against the channel estimation error and medium/low polynomial computational complexity.