PaperlayerView exact revision on arXiv

Source-linked AI summary

Coding the Beams: Improving Beamforming Training in mmWave Communication System

Y. Ming Tsang, Ada S. Y. Poon, Sateesh Addepalli

arXiv:1104.1007v4cs.NI

TL;DR

mmWave systems need efficient beam-pair discovery despite many antennas and few RF chains, while conventional in-packet training can cause large power variations. The paper proposes beam coding, which simultaneously steers coded beams, and reports flatter power behavior, no extra AGC resetting, and reduced training-packet size.

  • Problem

    In-packet BF training can produce large power variations across a packet, raising AGC and timing-synchronization challenges.

  • Method

    Beam coding simultaneously steers multiple highest-resolution beams and assigns each beam a unique signature sequence for receiver-side separation.

  • Results

    In NLOS, all beam-coding training data have γ < 1, whereas IEEE 802.11ad ratios range from 0 to 6; in LOS, 802.11ad reaches 14 for 16 beams.

  • Takeaways & Limitations

    Beam coding enables in-packet BF training without extra AGC resetting, supports use at every BF-training stage, and saves 4000 bits per trained beam direction.

Abstract

from arXiv · show

The mmWave communication system is operating at a regime with high number of antennas and very limited number of RF analog chains. Large number of antennas are used to extend the communication range for recovering the high path loss while fewer RF analog chains are designed to reduce transmit and processing power and hardware complexity. In this regime, typical MIMO algorithms are not applicable. Before any communication starts, devices are needed to align their beam pointing angles towards each other. An efficient searching protocol to obtain the best beam angle pair is therefore needed. It is called BeamForming (BF) training protocol. This paper presents a new BF training technique called beam coding. Each beam angle is assigned unique signature code. By coding multiple beam angles and steering at their angles simultaneously in a training packet, the best beam angle pair can be obtained in a few packets. The proposed BF training technique not only shows the robustness in non-line-of-sight environment, but also provides very flat power variations within a packet in contrast to the IEEE 802.11ad standard whose scheme may lead to large dynamic range of signals due to beam angles varying across a training packet.

I. INTRODUCTION

mmWave systems use many antennas and few RF chains to extend range while limiting hardware and power costs, making conventional MIMO unsuitable. The paper addresses efficient beam-pair discovery and proposes beam coding to support in-packet training with flatter received power.

  • Motivation: 60 GHz signals suffer 28 dB more free-space path loss than 2.4 GHz signals, motivating large antenna arrays for beamforming gain.The cited example notes that standards and some mmWave RFICs support 16 antennas.
  • Motivation: Only one analog chain is found in current mmWave RFIC designs because multiple chains are costly and GHz-rate sampling consumes substantial power.RF-applied antenna weights allow highly directional beam patterns with one analog chain.
  • Beam-pair discovery: BF training is required to discover the beam-direction pair with the highest channel gain because communication fails when devices point away from each other.The paper focuses on AP-device links but states that the protocol can also apply to device-to-device setup and other beam-pair metrics.
  • Existing training: Exhaustive beam-pair search is prohibitively slow, while multi-level and in-packet training reduce overhead by training multiple beams within one packet.In IEEE 802.11ad, in-packet training is used during beam refinement and can train multiple beams with the bandwidth of one packet.
  • Existing training: IEEE 802.11ad in-packet training can create large signal dynamic range and timing inaccuracies when beam directions change, restricting its use to beam refinement.The standard disallows AGC changes before each beam-direction change to preserve relative-delay information.
  • Beam coding: Beam coding steers multiple highest-resolution beams simultaneously, assigns each a unique signature sequence, and produces fairly uniform receive power across the packet.The scheme extracts channel information for all beam pairs while keeping covering beam directions unchanged throughout the packet.

II. IN-PACKET BF TRAINING

IEEE 802.11ad places beam refinement training inside packets after preamble and header coverage, using AGC and TRN fields to estimate channels for sequentially changed beams. The setup is illustrated with four steerable beams and one LOS plus one attenuated NLOS path.

  • Setup: The example restricts transmitter and receiver arrays to four beams, while noting that the setup generalizes to more beams.The channel example contains one LOS path and one NLOS path with attenuation a < 1.
  • Packet structure: IEEE 802.11ad steers preamble and header sections with low-resolution beams, then trains high-resolution beams in the BF training section.The BF training section contains AGC and TRN fields, with beam directions changed sequentially within the AGC field and one TRN field allocated per beam.

B. Exhaustive in-packet training

In-packet training makes exhaustive search feasible by reducing packet transmissions, while feedback training can complete faster but sacrifices full channel information and receive beamforming gain. These approaches expose a training-time versus communication-range tradeoff.

  • Exhaustive in-packet training: The example reduces exhaustive-search training packets from 16 to 4 by minimizing header and preamble overhead with in-packet training.The transmitter sends the same training packet four times while the receiver steers its beams one by one.
  • Feedback in-packet training: Feedback in-packet training can use piggybacked intermediate results, reducing explicit-feedback resource use during training.The association request can carry intermediate training results to improve training time.
  • Feedback in-packet training: Feedback training sacrifices full channel information and receive beamforming gain because only transmitter beams are trained initially and all receiver beams are active then.The paper identifies a tradeoff between training time and communication range and leaves it for future work.

III. BEAM CODING

Beam coding reduces in-packet power variation by steering multiple beams simultaneously and distinguishing them with signature-based antenna weighting. The scheme uses superimposed RF weights to form multiple directional maxima while preserving a stable packet-wide beam pattern.

  • Motivation: Beam coding steers multiple highest-resolution beams simultaneously across training fields to reduce receive-power variations and avoid explicit AGC resettling.The approach addresses the dynamic-range problem created when beam directions change within a packet.
  • Beam coding: Each beam receives a signature sequence so the receiver can identify signal strength and channel information for individual beam pairs.The selected training directions can also be designed to be orthogonal to further reduce power variation.
  • Assumptions: The model assumes omnidirectional antenna elements and a non-uniformly weighted linear array, while noting that uniform weighting is also feasible.The normalized inter-antenna distance is defined using the antenna spacing and signal wavelength.
  • RF implementation: Non-uniform antenna weights encode beams through individually adjusted RF antenna magnitudes and phases despite using a single analog chain.The paper notes that non-uniform weighting may provide the best performance and is supported by suitable circuitry.
  • Beam formation: Superimposing the weights for φ1 and φ2 creates a beam pattern with maximum points at both transmit directions.The same construction extends to four transmit beams and can be applied at the receiver.

B. Encoding the beams

Beam coding assigns known Walsh-derived signatures to simultaneously transmitted beams and recovers beam-pair strengths through receiver correlations. In the illustrated exhaustive scheme, four repeated frames complete training, while orthogonal coding helps preserve separability under multipath.

  • Signature assignment: Each transmit beam is assigned a unique signature sequence known to both transmitter and receiver.The signature sequences are derived from Walsh codes.
  • Weight construction: The transmitter computes antenna weights for consecutive CE fields using the beam responses and signature sequences.The construction is specified for four CE-field indexes in the illustrated procedure.
  • Training procedure: The transmitter sends the same packet 4 times while the receiver switches beams after every packet duration.This exhaustive beam coding scheme therefore takes 4 frames to complete training.
  • Beam-pair selection: Receiver correlations separate aligned beam pairs, yielding r(2,3) = 2 and r(1,4) = 2a; because a < 1, beam pair (2,3) is selected.The correlation comparison identifies the aligned pair with the larger value.
  • Robustness and scope: Walsh orthogonality and Golay channel-estimation sequences help preserve beam separation and recover relative delays and signal strengths across multipath taps.The paper also describes beam coding as a structure that could support compressed sensing, but does not pursue that benefit because training reduction is insignificant and receiver complexity increases.

C. Orthogonal beam directions

Beam coding reduces packet power fluctuation by encoding orthogonal beams, making transmission power independent of the signature codes used.

  • Training procedure: The exhaustive beam coding procedure trains receive beams across repeated packets to evaluate candidate beam combinations.The figure is identified as the exhaustive beam coding training procedure, with successive packets steering the receiver at different beams.
  • Orthogonal beam selection: Beam selections within a packet are required to be orthogonal to reduce power fluctuation.The procedure selects beams for coding and combines their antenna weights using additions and subtractions.
  • Orthogonality definition: Orthogonal beams are defined through zero inner product between their antenna-weight vectors.The definition normalizes each beam's antenna weights and uses the complex conjugate in the orthogonality condition.
  • Weight construction: The antenna weights for two coded beams combine their per-antenna weights with a signature-dependent plus or minus operation.The combined weight is expressed as 1/2(β_n(φ_i) ± β_n(φ_j)).
  • Power result: For orthogonal beams, normalized transmission power is |w|^2 = 1 and does not depend on the signature codes.Non-orthogonal beams instead produce nonzero cross terms that can change power across codes.

D. Uniformly weighted phased array

Uniformly weighted arrays can support beam coding, although multiple-beam steering increases sidelobes; sweeping all beams allows those sidelobe locations to be covered by other main lobes.

  • Array weighting: The beam coding scheme performs best with non-uniformly weighted arrays because it varies antenna-weight amplitudes.Uniform arrays adjust only weight phases, while non-uniform arrays support magnitude variations.
  • Sidelobe trade-off: Two-beam steering in a uniformly weighted ULA raises sidelobe level to -9 dB instead of -13 dB for single-beam steering.This is the stated sidelobe trade-off associated with simultaneous beam steering.
  • Training-stage acceptability: The increased sidelobes are acceptable during BF training because other swept beams eventually cover the sidelobe positions with their main lobes.The paper presents this coverage argument for the uniformly weighted array case.
  • Feasibility: For more than two simultaneously trained beams, Walsh signatures and an antenna count that is a power of 2 make beam coding analytically feasible in a uniformly weighted ULA.The paper states that simulations compare uniform and non-uniform arrays because the analytic proof is omitted.

IV. SIMULATIONS

The simulations compare beam coding with IEEE 802.11ad under the living-room channel model and evaluate receive-power flatness across a packet.

  • Power-flatness comparison: Beam coding provides comparably flat receive-power variations across a packet relative to IEEE 802.11ad.The comparison is performed in simulation as part of the evaluation of the proposed training scheme.
  • Simulation model: The simulations use the living room model from the IEEE 802.11ad channel document.Simulation parameters are reported in Tables I and II.

A. Receive power variations

The study evaluates receive-power variation using the power ratio γ under NLOS and LOS conditions, then quantifies beam coding’s training-packet savings. Beam coding keeps power variation lower than IEEE 802.11ad and removes the need for extra AGC fields.

  • Power-ratio metric: The power ratio γ compares training-section signal power with preamble-signal variation, using 3σ_prem as the reference scale.The 3σ_prem factor corresponds to an AGC range containing 99.5% of signals.
  • NLOS and LOS results: Under NLOS, all beam-coding training data has γ < 1, while IEEE 802.11ad ratios span 0 to 6.The receiver uses one antenna for a worst-case evaluation, and the transmitter varies the number of beams per packet.
  • NLOS and LOS results: In LOS, IEEE 802.11ad reaches a power ratio of 14 for 16 beams, whereas beam coding reaches 2.5.For beam coding, 95% of training data remains within power ratio 2, so no extra AGC resetting is required.
  • Training-packet overhead: Beam coding requires only the 1024-bit CE sequence, saving 4000 bits per beam direction compared with the IEEE 802.11ad training section.IEEE 802.11ad additionally uses AGC-resetting and relative-delay-estimation fields for each trained beam direction.

B. Uniformly weight linear array with phase quantization

The evaluation examines exhaustive beam coding with phase-quantized weights in a 16-antenna transmitter and receiver. Beam coding approaches exhaustive PbP performance despite coarse phase quantization, matching it in NLOS at three bits.

  • Evaluation setup: The simulation compares exhaustive beam coding with the exhaustive PbP scheme, which provides the upper-bound performance because it avoids multiple-beam steering.Both transmitter and receiver use 16 antennas.
  • Evaluation setup: Sixteen beam directions are trained simultaneously in each beam-coding packet, creating the worst-case phase-quantization condition.The setup intentionally trains all 16 directions at once to evaluate the most demanding beam-coding case.
  • Phase-quantization results: Beam coding approaches optimal exhaustive PbP performance even with two-bit phase quantization and matches it exactly in NLOS at three bits.With four-bit quantization, the beam-pattern plots show almost no distortion; two-bit quantization causes more distortion but preserves pointing directions.

V. CONCLUSION

The conclusion presents beam coding as an in-packet BF-training technique that stabilizes receive-power variation and reduces training-packet size. It also states that only mild mmWave RF-front-end modification is needed.

  • V. CONCLUSION: Beam coding enables in-packet BF training without extra AGC resetting by providing uniform receive-power variation while beam directions change within a packet.The conclusion states that this applies at every stage of BF training.
  • V. CONCLUSION: Beam coding greatly reduces the IEEE 802.11ad BF-training packet size by eliminating extra AGC and relative-delay-estimation fields.The conclusion connects the packet-size reduction to stable signal dynamic range during in-packet training.
  • V. CONCLUSION: The scheme requires no variable antenna-weight amplitudes, so supporting it needs only mild modification of the mmWave RF frontend.The conclusion identifies this as a hardware implementation property of beam coding.
Loading 1104.1007v4…