MU-MIMO: Non-Linear Precoding for DL-MIMO Document Number: IEEE C80216m-08_842 Date Submitted: 11th July 2008 Source: Laurent Mazet, Sheng Yang, Marc de Courville, Fred Vook(Motorola) Tsuguhide Aoki, Henning Vetter, Yong Sun (Toshiba) * Venue: Denver, CO, USA Purpose: Enabling non-linear predocing in DL MU MIMO SDD document. Notice: This document does not represent the agreed views of the IEEE Working Group or any of its subgroups. It represents only the views of the participants listed in the “Source(s)” field above. It is offered as a basis for discussion. It is not binding on the contributor(s), who reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE Patent Policy: The contributor is familiar with the IEEE-SA Patent Policy and Procedures: and. Further information is located at and.
2 Non-Linear Precoding motivation Principle –DL MU-MIMO scheme –Perform channel inversion or interference suppression at TX(BS) –Constrain transmitted energy through the use of a non linear perturbation Better performance than linear ZF/MMSE-SDMA –Limit RX(MS) complexity: through the use of non linear modulo arithmetic, the precoding/channel is transparent at the receiver providing that the non linear mapping is known at RX –Capture Dirty Paper Coding, Tomlinson-Harashima (TH) and Sphere Encoding (SE) precoding (vectorized THP) –Optimum scheme: i) achieves multi-user capacity; ii) maximizes the overall throughput; iii) cancels the inter user interference and iv) limits the total power at TX On the various perturbation based non linear constellation mapping: –Several mappings exist based on periodic replication of the original constellation by translation (original TH precoder) –Known issue: outer constellation points are inherently more sensitive to noise in the detection process induced by the “transparent” modulo decoder –A more robust scheme is possible: use of Periodically Flipped Constellation (PFC) Periodically replicate the original constellation on the integer lattice and apply up-down and left-right flips to every other image of the original constellation Apply associated modified modulo arithmetic for performing the decoding Same precoding complexity than traditional SE Classical constellation replication mapping results in detection error No detection error with periodically flipped constellation
3 NLP DL MU-MIMO simulations: SE vs. ZF/MMSE-SDMA Goal: compare ZF/MMSE(REG) SDMA with SE (traditional constellation mapping or PFC) Simulation parameters –SCME urban micro channel 2.5GHz, 10MHz bandwidth 10MHz (FFT size is 1024), 3kmph –Packet composed of 10 PUSC subchannels 2 frequency subchannels and 5 time slots –R=1/2 Conv. Code (o133, o171) –10 OFDM symbol delay between UL CSI estimation and DL, 2D MMSE channel estimator –Pilot pattern: 16e PUSC –4TX(BS), 4 1RX(MS) served –Normalized energy at TX(BS): not per user(MS) Performance –NLP outperforms LP: 6dB advantage at 5% PER for SE vs. MMSE-SDMA with higher diversity slope –Channel estimation incurs a 3dB shift at 3kmph for SE results with a perf. floor for SDMA –Additional 2dB gain for QPSK (0.5dB for 16QAM) for PFC vs. traditional SE constellation mapping Both SDMA and SE not viable at 120kmph
Impact of channel estimation impairment Linear precoding vs. Non-linear precoding 4 Non-linear precoding Linear precoding Configuration 4x(1,1,1,1) MU-MIMO configuration QPSK 1/2 rate conv. coding CNR is channel to noise power ratio in decibel Non-linear precoding outperforms linear precoding in any case of channel estimation impairments Linear precoding (LP) means Channel Inverse precoding, and nonlinear precoding (NLP) utilizes a Sphere Encoder in order to minimize the increase in transmit power.
Degradation on impairments of CSI estimation 5 Compared to perfect CSI estimation, linear precoding suffers much severe degradation on impairments of CSI estimation Non-linear precoding is much more robust than linear precoding With perfect CSI, 6 dB gain Higher gain obtained with worse channel estimation 12 dB gain at CNR = 18 dB
Discussions and recommendations MU-MIMO deployment is more suitable for low mobility; Linear precoding provides low complexity but with lower performance; Non-linear precoding achieves significant gain on higher MU-MIMO configuration; Non-linear precoding is much more robust to impairments of CSI estimation; m shall support both LP and NLP –LP might be suitable for 2x(1,1) MU-MIMO for simpler implementation but lower performance –NLP is more suitable for higher MU-MIMO configurations to achieve much higher system performance to meet IMT-advanced requirements. 6
7 Proposed Text for SDD