Sensing Aided OTFS Channel Estimation for Massive MIMO Systems

I. INTRODUCTION

• Pilot overhead in massive MIMO-OTFS scales with delay and Doppler spreads and transmitter antenna count, and can remain significant even with sparsity-based channel acquisition.The channel is typically sparse in delay, Doppler, and angle dimensions.Source: §I · p. 1
• The paper leverages radar sensing because delay-Doppler channel information is closely related to user and scatterer position, direction, and velocity.Its contributions include inferring path delay, Doppler, and AoD from radar signals and using those quantities in OMP-based sparse recovery.Source: §I · p. 1
• The evaluation framework combines co-existing wireless communication and radar sensing data to assess sensing-aided OTFS channel estimation.The proposed approach is reported to reduce pilot overhead while preserving NMSE performance.Source: §I · p. 1
• Simulation results show that the proposed sensing-aided OTFS channel estimation approach consistently outperforms conventional sparse recovery algorithms, achieving similar channel-estimation NMSE with 5 dB lower SNR and reducing pilot/channel acquisition overhead by more than 50% without NMSE degradation.Source: §I · p. 1

A. System Model

• The system model has a base station serving one highly mobile UE, with a ULA for communication and an FMCW radar whose environmental sensing data aids downlink channel estimation.The UE is assumed to have a single antenna for simplicity; multi-antenna UEs are described as an extension.Source: A. System Model · p. 2, Figure 1 · p. 2

B. OTFS Communication Signal Model

• The OTFS signal model maps delay-Doppler data into time-frequency and then delay-time signals, with cyclic prefixes added to avoid intersymbol interference.The adopted transmit window is rectangular, so its all-one windowing matrix can be omitted.Source: B. OTFS Communication Signal Model · p. 2

1) OTFS Demodulation:

• The received vector r is rearranged into a two-dimensional signal R before demodulation continues.Source: 1) OTFS Demodulation · p. 2
• The cyclic prefix is removed to obtain the delay-time receive signal, which is then converted to the time-frequency domain.Source: 1) OTFS Demodulation · p. 2
• The time-frequency receive signal is windowed and transformed to produce the delay-Doppler receive signal YDD.Source: 1) OTFS Demodulation · p. 2

C. OTFS Channel Model

• Each of L propagation paths is characterized by complex gain, delay, Doppler shift, and angle of departure, with delay and Doppler mapped to tap indices.The model assumes integer delay and Doppler cases for simplicity.Source: C. OTFS Channel Model · p. 3
• The received time-domain signal sums antenna-specific transmitted symbols shifted by delay taps and weighted by channel coefficients, with noise added.Source: C. OTFS Channel Model · p. 3
• The delay-Doppler input-output relation uses channel coefficients indexed by delay tap, Doppler tap, and transmit antenna.The channel coefficient HDD is associated with the corresponding delay, Doppler, and antenna indices.Source: C. OTFS Channel Model · p. 3

D. Radar Signal Model

• The base station uses an FMCW radar with one transmit antenna and B receive antennas, assuming sensing targets lie in the radar far field.The radar model and processing can be generalized to MIMO radars.Source: D. Radar Signal Model · p. 3
• Repeated identical chirps form a radar frame, and received signals are mixed with transmitted chirps, low-pass filtered, and represented as intermediate-frequency signals.The chirp frequency increases linearly from f0 to f0 + STc, and its effective bandwidth is Bw = STc.Source: D. Radar Signal Model · pp. 3–4
• The model neglects returns that interacted with multiple targets because they have smaller power, and its approximation holds when Sτ ≪ f0.Source: D. Radar Signal Model · p. 4
• The sampled signals from Nloop chirps and B receive antennas are collected into a radar data frame X.Each antenna has an independent receive chain with a mixer, low-pass filter, and ADC.Source: D. Radar Signal Model · p. 4

A. OTFS Frame Structure

• Fig. 3 depicts the adopted delay-Doppler frame with data, pilot, and guard symbols.Source: Figure 3 · p. 4, §III · p. 4
• Pilot and data symbols span the Doppler dimension in the same frame, while zero-power guard symbols separate them along delay.The guard length satisfies Mg ≥ Mmax so the data symbols do not interfere with the pilot symbols.Source: A. OTFS Frame Structure · p. 4
• The pilot overhead ratio is the quantity a pilot design and channel-estimation strategy seeks to minimize.Source: A. OTFS Frame Structure · p. 4

B. MIMO-OTFS Downlink Channel Estimation: A Sparse Recovery Formulation

• Rearranging pilot symbols, received pilots, and channel coefficients yields a matrix-vector sparse recovery problem with sensing matrix Ψ = Z ⊙ P.Each element of h corresponds to a delay-Doppler tap.Source: B. MIMO-OTFS Downlink Channel… · pp. 4–5
• Transforming the transmit-antenna domain to the virtual angle domain further increases sparsity, producing the angle-domain formulation ˜Ψ = ΨA^H and ˜h = Ah.The delay-Doppler channel is sparse because only a limited number of propagation paths and near-zero Doppler taps have relatively high power.Source: B. MIMO-OTFS Downlink Channel… · p. 5
• The sparse formulations can be solved with conventional compressed-sensing algorithms, including basis pursuit and matching pursuit.Source: B. MIMO-OTFS Downlink Channel… · p. 5

IV. SENSING AIDED DELAY-DOPPLER CHANNEL ESTIMATION

• Radar sensing at the base station is proposed to aid MIMO-OTFS channel estimation, using environmental information that is relevant to delay-Doppler communications.Source: §IV · p. 5

A. Key Idea: Sensing Aided Delay-Doppler Communications

• UE position and velocity and reflector geometry can inform potential path delay, Doppler, and AoD/AoA, with possible uses in channel estimation, feedback, and resource allocation.The paper notes radar’s potential availability through shared communication hardware and software, its direct relation to wireless channels, and its ability to obtain NLoS sensing information.Source: A. Key Idea: Sensing Aided… · p. 5

B. Relation Between Communication and Radar Channels

• For backscattering, radar-path delays and Doppler velocities are approximately twice those of corresponding communication paths, while their angles are approximately the same; other mismatched paths are not handled.Source: B. Relation Between Communication and… · p. 6

C. Radar Processing

• Radar processing removes static clutter, applies range, Doppler, and angle DFTs, then uses CFAR peak detection to extract propagation delay, Doppler velocity, and AoA.The processing uses 2D-CFAR on a range-angle heatmap, followed by 1D-CFAR on Doppler dimensions, with non-maximum suppression.Source: C. Radar Processing · pp. 6–7, Figure 5 · p. 6

D. Sparse Recovery with Radar Sensing Information

• Radar-detected delay, Doppler velocity, and AoA are converted to delay, Doppler, and angle-domain channel indices to form the radar-informed support set S_r.The delay normalization assumes the shortest radar path is detected as a peak, considered reasonable because it is likely among the strongest paths.Source: D. Sparse Recovery with Radar Sensing… · p. 7
• The support set S_r is sent to the UE over the control channel for use in sparse channel recovery.Source: D. Sparse Recovery with Radar Sensing… · p. 7
• The modified OMP initializes estimated support with radar-derived indices and adaptively sets its maximum iterations and support size according to the number of detected peaks.The stated intuition is that more potential reflectors make more resolvable communication paths likely, while radar tends to detect more peaks.Source: D. Sparse Recovery with Radar Sensing… · pp. 7–8

V. SIMULATION DATA GENERATION

• The ray-tracing scenario models a downtown area with one horizontal and two vertical streets, buildings, and a base station along the horizontal street.The UE is randomly placed within the designated area and assigned a velocity uniformly sampled from 50 to 90 m/s.Source: §V · p. 8
• Across 100 scenes with different UE locations and velocities, ray tracing generates path complex gains, delays, Doppler frequencies, AoA, and AoD for communication and radar channels.These parameters are used to obtain the communication channel HDD and radar data frame X.Source: §V · p. 8
• The communication and radar channels use 6 GHz and 28 GHz frequency bands, respectively.Source: §V · p. 8

• The proposed sensing-aided channel estimation is compared with conventional OMP using NMSE and pilot overhead ratio.Source: §VI · p. 8
• Figure 7 reports NMSE performance under different SNRs with 32 BS antennas and a 20% pilot overhead ratio.Source: Figure 7 · p. 8

A. Does Sensing Improve Channel Estimation Accuracy?

• The proposed approach achieves similar NMSE with 4dB lower SNR than OMP (angle domain), with a slightly larger gap at low SNR.The passage attributes greater robustness at high noise to using channel support extracted from radar signals.Source: A. Does Sensing Improve Channel… · p. 8

B. Does Sensing Reduce the Pilot Overhead?

• Across pilot overhead ratios, the proposed approach consistently outperforms conventional OMP (angle domain) in NMSE.Source: B. Does Sensing Reduce the Pilot… · pp. 8–9
• At η = 0.15, the proposed approach achieves similar NMSE to OMP (angle domain) at η = 0.3, a 50% reduction in pilot overhead.Source: Figure 8 · p. 9, B. Does Sensing Reduce the Pilot… · p. 9

VII. CONCLUSIONS

• The method formulates downlink massive MIMO-OTFS channel estimation as sparse recovery aided by radar sensing, evaluated with co-existing radar and communication data in accurate 3D ray tracing.Source: §VII · p. 9