PaperlayerView exact revision on arXiv

Source-linked AI summary

Directional Initial Access for Millimeter Wave Cellular Systems

C. Nicolas Barati, S. Amir Hosseini, Marco Mezzavilla, Sundeep Rangan, Thanasis Korakis, Shivendra S. Panwar, Michele Zorzi

arXiv:1511.06483v2cs.ITcs.NI

TL;DR

MmWave initial access must discover a cell and establish directional links despite a potentially large beam-search space and the resulting control-plane delay. The paper compares synchronization and random-access designs across scanning, signaling, and receiver architectures, finding that low-resolution fully digital beamforming can substantially reduce access delay and overhead relative to sequential analog scanning.

  • Problem

    MmWave initial access must discover a cell, establish a link, and determine suitable UE and BS beam directions despite a potentially large angular search space.

  • Method

    The paper analyzes five initial-access design options, deriving delay and overhead behavior for synchronization and random access under analog and low-resolution fully digital beamforming.

  • Results

    Low-resolution fully digital beamforming significantly outperforms the other schemes across overhead ratios and can reduce access delays to a few milliseconds.

  • Takeaways & Limitations

    Fully digital reception can remove sequential scanning delay and multiplex additional random-access channels, while low-resolution quantization can limit added power consumption.

Abstract

from arXiv · show

The millimeter wave (mmWave) bands have recently attracted considerable interest for next-generation cellular systems due to the massive available bandwidths at these frequencies. However, a key challenge in designing mmWave cellular systems is initial access -- the procedure by which a mobile establishes an initial link-layer connection to a base station cell. MmWave communication relies on highly directional transmissions and the initial access procedure must thus provide a mechanism by which initial transmission directions can be searched in a potentially large angular space. Design options are compared considering different scanning and signaling procedures to evaluate access delay and system overhead. The channel structure and multiple access issues are also considered. The analysis demonstrates significant benefits of low-resolution fully digital architectures in comparison to single stream analog beamforming.

I. INTRODUCTION

The paper frames mmWave initial access as a control-plane latency problem complicated by directional beam discovery. It compares design options for synchronization and random access, including low-resolution fully digital architectures that can inspect multiple directions simultaneously.

  • Motivation: mmWave initial access discovers a potential cell and establishes a link-layer connection, especially after intermittent links or idle-mode recovery.
  • Motivation: Directional transmissions require the UE and BS to determine suitable beamforming directions, creating a potentially large angular search space.
  • Contributions: The procedure retains LTE-like basic steps but modifies synchronization and random access so both sides learn initial beamforming directions.
  • Contributions: The paper evaluates the whole initial-access procedure across five design options, measuring delay as a function of system overhead.
  • Contributions: The study compares analog beamforming with a theoretical low-resolution fully digital architecture that can look in all directions at once.
  • Scope: The standalone mmWave evaluation uses a realistic LOS/NLOS channel model while assuming that establishing NLOS links is achievable.

B. Synchronization and Random Access Signals

The synchronization and random-access design uses periodic narrowband signals and adapts LTE-like stages to directional mmWave operation. Design choices determine the angular search size and the resulting delay-overhead tradeoff.

  • Signal structure: Synchronization and random-access signals are modeled as narrowband waveforms transmitted periodically, with signal duration and period treated as shared parameters for comparison.
  • Procedure: The mmWave procedure modifies LTE-like steps so the UE learns its beam direction during synchronization and the BS learns its direction during random access.
  • Design options: The study forms five analyzed designs from choices for synchronization transmission and synchronization/random-access reception, although eight combinations are possible.
  • Sequential scanning: Sequential analog beamforming cycles through L transmitter-receiver direction pairs, so one scan cycle lasts L Tper seconds.
  • Sequential scanning: The angular search size is L = NtxNrx when both ends scan, L = Ntx when only the transmitter scans, and L = Nrx when only the receiver scans.
  • Digital reception: Digital reception can set L = 1 while preserving receive directivity because it can inspect all Nrx directions simultaneously.

III. SIGNAL DETECTION UNDER A SEQUENTIAL BEAMSPACE SCANNING

The detector models repeated beamspace transmissions and tests whether a signal is present while identifying its strongest direction. Analysis uses an idealized aligned-channel model, whereas simulations use realistic measurement-based channels.

  • Detection model: The receiver observes K scan cycles, each containing L transmissions, and detects whether a signal is present while determining its arrival direction.
  • Signal model: Each received transmission is represented in an M-dimensional signal space, with M approximately equal to TsigWsig.
  • Channel assumption: The channel model assumes a single line-of-sight path exactly aligned with one of the L transmitter-receiver beamspace directions.
  • Channel assumption: The scalar channel coefficient may vary across subsignals, transmissions, and scan cycles, while the simulations use a realistic measurement-based channel model.
  • GLRT detection: Unknown signal and noise parameters are handled with a generalized likelihood ratio test that compares the signal-absent and signal-present hypotheses.
  • GLRT detection: The implementation computes matched-filter correlations, selects the optimal beamspace direction, and uses the resulting likelihood difference in the hypothesis test.

B. Hybrid and Digital Beamforming

The analysis compares analog, hybrid, and digital beamforming for mmWave initial access, focusing on synchronization and random-access delay. Low-resolution digital reception can examine many directions in parallel and substantially reduce delay relative to analog beamforming.

  • Hybrid Beamforming: Hybrid reception obtains S power measurements per time instant, accelerating detection by a factor of S over a single analog beamformer.The detector remains identical to the analog case, but parallel RF chains enable simultaneous measurements.
  • Digital Beamforming: Digital reception tests all Nrx beamspace directions at each time step, providing greater acceleration than analog beamforming.The analysis assumes arrival angles align with beamspace directions, although digital reception could also test intermediate angles.
  • Delay Metrics: The paper evaluates synchronization delay, random-access preamble detection delay, and their sum as the access delay governing the initial-access latency baseline.These first two phases dominate the delay before later beam fine-tuning and channel estimation.
  • Analysis Model: The synchronization analysis targets UEs whose SNR exceeds γtgt and models the received signal using antenna gains Grx and Gtx, with synchronization gain Gsync_tx.Correct beam alignment occurs once per scan cycle, while synchronization overhead depends on Tsig,sync and Tper,sync.
  • Analog Beamforming: For analog beamforming, delay grows linearly with the combined gain GrxGtx when synchronization duration is unconstrained.With a minimum practical signal duration, omni-directional transmission has a lower bound of Tmin_sig,sync, while directional transmission is lower-bounded by GsyncTmin_sig,sync.
  • Digital Beamforming: Digital beamforming at the UE improves delay over both analog options by a factor of Grx, especially when the UE has high directional antenna gain.The analysis therefore identifies a substantial delay benefit for fully digital reception in high-gain UE settings.

A. System Parameters

The paper combines simplified analytical delay models with simulations using realistic signal-processing and channel assumptions. The simulation framework accounts for detector search hypotheses, channel coherence constraints, and measurement-based path-loss modeling.

  • Analysis Scope: The simplified analysis provides intuition about delay but cannot by itself support detailed system design choices.The paper therefore evaluates delay using realistic system parameters and channel models.
  • Signal Parameters: Signal duration and bandwidth are selected so each Tsig×Wsig subsignal region is approximately flat according to channel coherence time and bandwidth.Tsig is also chosen short enough to limit Doppler variation across a subsignal.
  • Detection Model: The detector search accounts for false alarms across signal, delay, and frequency-offset hypotheses, with Ndly representing delay hypotheses within each transmission period.The delay grid searches sample positions between successive transmission periods.
  • Simulation Parameters: Tables III and IV provide the default simulation parameters and the SNR-specific simulation parameters, respectively.These parameter tables define the settings used unless the analysis states otherwise.

B. SNR

The SNR analysis models a 100 m mmWave cell with LOS/NLOS links, no interference, and total-bandwidth omni-directional SNR. It also evaluates low-resolution fully digital reception, finding limited quantization loss at 3 bits.

  • SNR model: The model places the mmWave BS at the center of a 100 m-radius cell and randomly drops UEs to compute path loss.Links are selected as LOS or NLOS using distance-dependent probabilities; outage links are ignored because they cannot establish communication.
  • SNR model: The path-loss model is PL = α + 10β log 10(d) + ξ [dB], with ξ ∼N(0, σ2), and parameters determined by LOS or NLOS conditions.
  • SNR assumptions: The analysis ignores intercell interference and in-cell or out-of-cell uplink random-access collisions under synchronization and non-overlapping-frequency assumptions.
  • SNR results: SNR distributions are plotted for downlink and uplink over Wtot = 1 GHz using omni-directional SNR without beamforming gain.The figure includes 1% and 5% percentiles and median lines; total-bandwidth SNR defines the operating regime and data rates.
  • SNR interpretation: Synchronization and random-access signals have higher SNR in their narrower signal bandwidths than the total-bandwidth SNR.The paper uses total-bandwidth SNR because narrowband signal SNR is meaningful only in the context of those signals.
  • Low-resolution digital BF: Fully digital receiver options require one ADC per antenna element, creating a potential power-consumption cost that is especially important at the UE.
  • Low-resolution digital BF: 3-bit ADC resolution limits quantization loss to 0.15 dB in the cited model.ADC power generally scales as 2^b, and the analysis constrains fully digital architectures to at most 3 bits per I/Q dimension.

D. Multipath NYC Channel Model

The paper evaluates synchronization scanning and delay for several mmWave initial-access designs, including a single-path detector assumption and low-resolution digital reception. Digital reception can inspect multiple directions simultaneously and significantly outperforms the alternatives across overhead ratios.

  • Channel and detector assumptions: The detector analysis assumes channel gain is concentrated in one path aligned with one of L beamspace direction pairs.The paper identifies angular-space size and scan time, rather than multipath itself, as the fundamental difficulty in Sync/RA detection.
  • Synchronization scanning: The number of synchronization scan cycles K is estimated from Monte Carlo PMD curves versus SNR for DDD/DDO designs.The simulations use a false-alarm rate of RFA = 0.01 per scan cycle and compare the curves with low-percentile downlink SNRs.
  • Synchronization scanning: Users above the relevant SNR threshold can meet the detection target in one scan cycle.
  • Synchronization delay: Synchronization delay is computed as Dsync = KLTper,sync = KLTsig,sync / φov,sync and evaluated at fixed overhead φov,sync = 5%.Table VI reports detection delays in milliseconds and includes the angular spaces scanned for synchronization and random access.
  • Synchronization signaling: The minimum synchronization signal duration considered is Tsig,sync = 10 µs, yielding 10 degrees of freedom when Wsig = 1 MHz.Nine degrees of freedom remain for noise-variance estimation.
  • Design comparison: For analog beamforming, omni-directional transmissions outperform directional transmissions, while ODigDig significantly outperforms other schemes across overhead ratios.The digital gain comes from looking in multiple directions simultaneously while retaining directionality at both receiver and transmitter, using low-resolution quantization.

F. Random Access Delay

Random access delay and overhead depend strongly on beamforming architecture and signaling assumptions. Digital reception consistently offers the best delay–overhead tradeoff, while frequency multiplexing can further reduce overhead.

  • The analysis conservatively applies the same random-access overhead to digital and analog beamforming because potential multiplexing gains are deferred.The omitted gain may require dynamic range unavailable with low-resolution ADCs.
  • After synchronization, the UE already knows the BS angle, so random access searches a smaller beam space and fewer delay hypotheses.The UE transmits its random-access preamble only in the learned direction.
  • Digital RX schemes outperform analog and mixed schemes in random access delay and overhead, in some cases by orders of magnitude.The comparison covers both edge- and high-SNR regimes.
  • Digital beamforming can receive random-access requests and connected-user data simultaneously on different frequencies, reducing overhead beyond the conservative calculation.This multiplexing is unavailable to single-stream analog beamforming, which examines one direction at a time.
  • Fully digital architectures remove sequential scanning at the receiver and can reduce access delay to a few milliseconds.The receiver can inspect all directions at once, while digital beamforming also supports channel multiplexing during random access.

APPENDIX A DERIVATION OF THE GLRT

The appendix derives the generalized likelihood-ratio detector by exploiting the conditional Gaussian signal model and minimizing the negative log likelihood over its unknown parameters.

  • The received signal is conditionally Gaussian given the signal parameters, allowing the detector to be formulated through a negative log likelihood.
  • The detector uses minimum log likelihoods after applying the parameter-specific minimizations.
  • The correlation quantity ρkℓ is defined through the earlier signal-model expression in (8).
  • The derivation assumes noise vectors are independent across different transmissions.
  • The derivation minimizes the likelihood over nuisance parameters including bℓ0, τkℓd, and ψkd.The appendix gives separate minimization steps for delay and signal-related parameters.

APPENDIX B SIMULATION PARAMETER SELECTION DETAILS

The simulations select signal, antenna, false-alarm, and timing parameters to balance detection quality, channel coherence, overhead, and practical system considerations.

  • The simulations use Ndiv = 4 narrow-band subsignals, providing a stated tradeoff between frequency diversity and coherent combining.
  • Signal durations of 10, 50, and 100 µs are evaluated, with the selected durations intended to remain within channel-coherence conditions at mmWave frequencies.The text gives a 30 km/h UE at 28 GHz as an example and notes delay spreads below 30 ns.
  • Signal transmission period trades off synchronization delay against overhead: for Tper = 1 ms, durations of 10, 50, and 100 µs produce 1%, 5%, and 10% overhead.More frequent transmissions support tracking and resynchronization but increase overhead.
  • The false-alarm target is RFA = 0.01 false alarms per transmission period, with per-hypothesis thresholds determined from Nhyp = NsigNdlyNFO.
  • The synchronization phase searches delays over [0, Tper], yielding Ndly = 10^4 for Wsig = 1 MHz and Tper = 5 ms.Random access uses a smaller delay-hypothesis count because it follows synchronization and depends on propagation round-trip time.
  • The baseline signal counts are Nsync = 3 and Nrach = 64, matching corresponding LTE values while allowing denser deployments to require more synchronization signals.
  • The appendix reports a random-access false-alarm expression evaluating to 5.1079(10)^−6.
  • The antenna assumptions use 8 × 8 elements at the BS and 4 × 4 elements at the UE, with half-wavelength spacing.
Loading 1511.06483v2…