华为主推的Polar Code提案内容

华为主推的Polar Code提案在3GPP会议中通过了，这几天媒体刷爆了这件事，从最初的“碾压”、“完爆”到现在的“误读”，科技界的事倒像是娱乐圈的事。无论媒体人怎么报道，我想很多人还是想知道这个提案到底是什么。由于最近工作关系即将变动，目前手头上已经没有工作的事情，刚好有空来翻翻这篇提案的一些细节。

2016年10月，华为/海思在葡萄牙首都里斯本，以PPT文档的形式（编号R1-1610667）给出了几种信道编码方案的比较。文档从性能、灵活性、实现的复杂度、时延这几个方面对比了Polar、LDPC、Turbo、TBCC等几种编码方案的特点，原文如下文的图1所示。

（图1 华为/海思2016年10月份的原始提案）

从这篇原始文档中可以看到，关于编码方案到底采用哪一种，实际在之前的几次会议中已经有过讨论，但没有定论，可见一篇提案写进标准规范不仅仅是脑力上的付出，也是要拼体力的。

图1文档的关键点转换成下文的表1更方便阅读。从表1中可以看到，华为/海思除了基于自己公司的研究成果外（R1-1608864等），还参考了众多其他同行的研究成果，比如中兴通讯（R1-166411等）、展讯（R1-1608922等）、诺基亚（R1-1609583等）、电信研究院（R1-1609578等）、Intel（R1-167703等）、MTK（R1-1609336等）等等。可以看到，这篇提案（其它很多提案都类似）集合了众多厂家的研究成果，很难说归属于某一家所有，科学是没有国界的。

（表1）

R1-1610667 Way Forward on Channel Coding Observations Huawei, HiSilicon Proposed observations: Performance: • Polar vs. LDPC o For list 32 decoder for Polar, Polar has better BLER performance compared to LDPC for the evaluated info block lengths [Huawei R1-1608864, ZTE R1-166411, Spreadtrum R1-1608922] o For list 8 decoder for Polar, Polar has better BLER performance compared to LDPC for the evaluated info block lengths/code rates except for high coding rates and large info block size [Huawei R1-1608864] o For list 4 decoder for Polar, Polar has better BLER performance compared to LDPC for small info block [Nokia R1-1609583]  • Polar vs. LTE Turbo o For list 32 decoder for Polar, Polar has better BLER performance compared to Turbo for the evaluated info block lengths [Huawei R1-1608864, Huawei R1-164377, QC R1-164700, ZTE R1-166411, Spreadtrum R1-1608922, DCM R1-1610060, CATR R1-1609578] o For list 8 decoder for Polar, Polar has better BLER performance compared to Turbo for the evaluated info block lengths and coding rates except rate 1/3 [Huawei R1-1608864, DCM R1-1610060] o For list 4 decoder for Polar, Polar has better BLER performance compared to Turbo for small info block [Nokia R1-1609583] • Polar vs. TBCC o For list 32 decoder for Polar, Polar has better BLER performance compared to TBCC for info block lengths up to 200 bits [Huawei R1-1608863, ZTE R1-1608973, CATT R1-1608771, Intel R1-167703, CATR R1-1609579] o For list 4 decoder for Polar, Polar has better BLER performance compared to TBCC for info block [Nokia R1-1609589, Nokia R1-1609594] Flexibility: • For Polar, 1-bit granularity can be achieved for all coding rates and for full range of block size o [Huawei R1-167209],[Intel R1-164184] • For LDPC, 1-bit granularity cannot be achieved with reasonable implementation complexity, due to multiple PCMs needed to support all coding rates with a complex switch network   o [Samsung R1-166771],[Qualcomm R1-1610137],[Nokia R1-1609584],[Samsung R1-1609067] • For Turbo, 1-bit granularity is feasible for all coding rates and for full range of block size  Implementation complexity: • For polar, list 8 decoder is deemed implementable for all info block size and list 32 is deemed implementable for small info block size o [Huawei R1-1608865] [MediaTek R1-1609336] • For LDPC, LOMS/LNMS decoder with iterations up to 25 is deemed implementable for info block granularity coarser than LTE o [Qualcomm R1-1610139],[MediaTek R1-1609337],[Nokia R1-1609584],[Samsung R1-1609067] • For Turbo, decoder complexity is impractical for large info block size Latency: • For Polar o with list 8 decoder, a decoding latency of 16us can be met for all info block length  • [Huawei R1-1608865]  o with list 32 decoder, a decoding latency of 1.7us can be met for small info block length (80bit)   • [Huawei R1-1608865] • For LDPC, a latency of 16us can be met for all info block length o [Qualcomm R1-166372]  • For Turbo, a latency of 16us can be met for small info block length

经过一系列的讨论研究，2016年11月，在美国Reno举行的3GPP例行会议中正式通过了这一提案，文档编号为R1-1613211，如图2所示。需要注意的是，这项提案的通过，不仅仅是华为/海思的努力，如果没有众多国内外同行的支持也是不行的。值得一提的是，在这份支持的名单中，国内公司除了中兴通讯、移动、电信这类巨头外，还出现了酷派、海能达、展讯、努比亚、OPPO、小米这类发展迅速的企业，这些站在华为背后的男人们，是一个亮点。

（图2 华为/海思主导的Polar Code方案获得通过）