A high-speed QR decomposition processor for carrier-aggregated LTE-A downlink systems

Abstract

Abstract-This paper presents a high-speed QR decomposition (QRD) processor targeting the carrier-aggregated 4 × 4 Long Term Evolution-Advanced (LTE-A) receiver. The processor provides robustness in spatially correlated channels with reduced complexity by using modifications to the Householder transform, such as decomposing-target redefinition and matrix real-valued decomposition. In terms of hardware design, we extensively explore flexibilities in systolic architectures using a high-level synthesis tool to achieve area-power efficiency. In a 65 nm CMOS technology, the processor occupies a core area of 0.77 mm 2 and produces 72 MQRD per second, the highest reported throughput. The power consumed in the proposed processor is 127 mW.

Energy Efficient Group-Sort QRD Processor With On-Line Update for MIMO Channel Pre-Processing

“…To ease the discussion we divide the table into two sections (columns), i.e., application-specific integrated circuit (ASIC) and ASIP, and cover 3 broad features (rows). Note that the 8 8 real-valued matrix supported by [11] and [3] in Table VII is an equivalent real domain transformation of the 4 4 complex-valued matrix. Therefore, it is fair to compare system performance with these works.…”

“…For ASIC implementations, such as in [11] and [20], wordlength optimizations are performed to trade-off between hardware cost and performance. Their results show that QR pre-processing for LTE systems can be performed accurately using around 16 bits.…”

“…Specifically, in cases where column changes take place at the right-most of , only one element in the lower triangular part of (i.e., , 4th row 3rd column of ) becomes non-zero. This implies that triangularization of can be significantly simplified by nulling the single non-zero element instead of operating on all columns afresh as (7) where (8) In (8), is the complex conjugation, and are defined as (9) After triangularizing , exact and in the proposed QRU scheme are obtained, expressed as (10) (11) Note that the lower-right diagonal element of in (11), i.e., , has been transformed from real to complex-valued domain during this exact QR update. This can be easily resolved by performing an additional real-domain transformation, e.g., using Givens rotation matrix [3], in case if real-valued diagonal elements are required.…”

“…QRD of a matrix can be implemented efficiently using different algorithms, such as Givens rotation [10], Householder [11], Gram-Schmidt [12], by exploiting the intricate mechanisms of each of these algorithms. However, the computation involved is expensive and has to be performed for each sub-carrier.…”

Efficient MIMO Preprocessor With Sorting-Relaxed QR Decomposition and Modified Greedy LLL Algorithm

“…The majority of existing QRD/SQRD algorithms are based on four basic methods: the Householder transformation (HT) method, Givens rotation (GR) method, Gram-Schmidt (GS) method, and Cholesky method. HT method [12], [13] has a good adaptability even for decomposing correlated matrices, but relatively high complexity makes HT method unfavorable for hardware implementation. GS method [14], [15] is notable for its low latency because it decomposes matrices at columnwise.…”