Virtual DPD Neural Network Predistortion for OFDM-based MU-Massive MIMO

Abstract

The nonlinearities of power amplifiers in massive MIMO arrays introduce unwanted spectral regrowth, which is typically avoided via digital predistortion at each amplifier. However, as the number of base station antennas scales up, so does the computational burden of per-antenna linearization. This work introduces a neural-network virtual digital predistortion (vDPD) scheme that operates before the linear precoder for OFDM-based massive MU-MIMO systems. By applying predistortion before the precoder, complexity scales primarily with the number of users. We can achieve comparable linearization along the user beams by training our neural network based on the memory polynomial, predistortion-per-antenna approach. We verify our algorithm through an exhaustive simulator that includes highorder amplifier nonlinearities, memory effects, and variance across the amplifier models.

Frequency-domain digital predistortion for Massive MU-MIMO-OFDM Downlink

“…where the DPD output u PA b,n corresponds to the PA input in (7). In the case of massive MU-MIMO, one can use one DPD for each PA.…”

“…To address the complexity problem of TD DPD in massive MU-MIMO, the authors in [7] proposed a NN-based DPD, which operates in the FD prior to the precoder. In this paper, we refer to this FD NN-based DPD model as FD-NN.…”

“…These methods unavoidably degrade the linearization performance as each PA has a different behavior due to variations in component characteristics. Another research direction goes toward frequency domain (FD) DPD [6], [7], where DPD is implemented before the OFDM inverse discrete Fourier transform (IDFT). The complexity of the DPD is reduced as the DPD sampling rate becomes equal to the symbol rate, instead of using oversampling as in conventional DPD methods.…”

“…The complexity of the DPD is reduced as the DPD sampling rate becomes equal to the symbol rate, instead of using oversampling as in conventional DPD methods. While [6] considers hybrid massive MIMO, [7] considers a fully digital MU-MIMO system with a neural network (NN)-based DPD, where the DPD is implemented prior to the precoder. Thus, the complexity of the DPD is reduced as the dimensionality of the signal increases with the number of UEs instead of the number of BS antennas.…”

“…However, this method requires several inverse discrete Fourier transforms (IDFTs) and discrete Fourier transforms (DFTs) as the NN-based DPD still operates in the time domain (TD). More problematically, the complexity cost of the method in [7] still increases with the number of BS antennas due to the additional precoding cost of the guard-band subcarriers. To address this complexity problem, an FD DPD model that only processes data subcarriers without IDFTs would be a possible solution.…”