Sounding Reference Signal (SRS)

In Time Division Duplex (TDD) systems, the uplink and downlink share the same frequency band. The SRS exploits channel reciprocity: the gNB assumes the channel estimated from the UE's SRS transmission is identical to the downlink channel. This allows the gNB to calculate downlink precoding weights directly without requiring the UE to measure and report CSI, eliminating massive feedback overhead. The accuracy of this assumption depends on channel calibration to compensate for mismatched transmit/receive chains.

5G NR supports SRS transmission from 1, 2, or 4 antenna ports on the UE. Multi-port SRS enables uplink MIMO channel estimation, allowing the gNB to determine the full spatial matrix between the UE's transmit antennas and the gNB's receive antennas. This is critical for:

• Codebook-based uplink precoding: The gNB selects the optimal precoding matrix for the UE's PUSCH.
• Antenna switching: A UE with fewer transmit chains than receive antennas can switch its transmitter across antennas to sound the full downlink channel for reciprocity-based beamforming.

An SRS resource set groups one or more SRS resources with a common usage: beam management, codebook-based, non-codebook-based, or antenna switching. Each resource is associated with a spatial relation to another reference signal (e.g., SSB, CSI-RS, or another SRS). This spatial relation tells the UE which transmit beam to use, ensuring the SRS is transmitted in the correct spatial direction for the gNB to measure the intended beam pair.

To sound a wide bandwidth with a power-limited UE, SRS supports frequency hopping. The UE transmits SRS on a subset of subcarriers in one OFDM symbol, then hops to a different frequency region in the next SRS symbol. Over multiple hops, the gNB can reconstruct a wideband channel estimate. 5G NR defines hopping patterns at the slot and symbol level, balancing sounding bandwidth against latency and overhead.

Beyond channel sounding, SRS is a fundamental uplink positioning reference signal. Multiple gNBs (or TRPs) measure the Time of Arrival (ToA) and Angle of Arrival (AoA) of the UE's SRS. Techniques include:

• UL-TDOA: Hyperbolic positioning via time difference of arrival at multiple gNBs.
• UL-AoA: Triangulation using angle measurements. This enables sub-meter accuracy in 5G NR positioning, critical for Industry 4.0 and autonomous guided vehicles.

Neural networks are being applied to SRS processing to overcome classical limitations:

• SRS Channel Prediction: Recurrent or transformer networks predict the channel state for future slots, compensating for channel aging in high-mobility scenarios.
• Super-Resolution SRS: Deep learning reconstructs a wideband channel estimate from a sparse or partial SRS sounding, reducing pilot overhead.
• Direct Precoding Inference: A neural network maps raw SRS measurements directly to optimal downlink precoding matrices, bypassing explicit channel estimation.

The physical principle that makes SRS viable for downlink precoding. In Time Division Duplex (TDD) systems, the uplink and downlink share the same frequency band, so the channel response measured on the SRS is assumed identical to the downlink channel.

• Requires calibration of transmit/receive RF chains to compensate for hardware asymmetries
• Enables the gNB to derive downlink CSI without UE feedback
• Fundamentally different from FDD systems, which require explicit CSI reporting via CSI-RS

The 3GPP NR standard defines how SRS transmissions are structured in time, frequency, and spatial domains. Key parameters include:

• SRS Resource Set: A collection of SRS resources with a common usage type (codebook, non-codebook, antenna switching, or beam management)
• SRS Ports: Each resource can be configured with 1, 2, or 4 antenna ports for multi-layer sounding
• Comb Structure: Comb-2 or Comb-4 defines the frequency-domain interleaving pattern, allowing multiple UEs to sound simultaneously on orthogonal subcarriers
• Periodicity: Configurable from 1 ms to 320 ms, balancing channel aging against uplink overhead

The gNB uses SRS measurements to compute the optimal downlink precoding matrix. The process involves:

• Channel matrix estimation from the received SRS across all antenna ports
• Singular Value Decomposition (SVD) to determine the dominant eigenmodes of the channel
• Zero-Forcing or MMSE precoding to spatially separate users and null inter-user interference
• The accuracy of this precoding directly depends on SRS signal-to-noise ratio and channel coherence time

These two reference signals serve complementary roles in 5G NR channel estimation:

• SRS (Uplink): Transmitted by the UE, used by the gNB to estimate the uplink channel. In TDD, this is repurposed for downlink precoding via reciprocity
• CSI-RS (Downlink): Transmitted by the gNB, measured by the UE to report CQI, PMI, and RI back to the network
• SRS enables implicit feedback-free operation in TDD, while CSI-RS supports explicit feedback in FDD
• SRS can also be used for uplink link adaptation and positioning in NR Release 16+

The temporal validity of SRS-based channel estimates is bounded by the channel coherence time. When the UE moves or the environment changes between SRS transmissions, the precoding becomes stale.

• Doppler spread determines the maximum SRS periodicity before performance degrades
• At 3.5 GHz with pedestrian speeds, coherence time is ~10-20 ms
• Predictive algorithms, including Kalman filters and recurrent neural networks, can extrapolate CSI between SRS occasions
• Trade-off: shorter SRS periodicity improves accuracy but increases pilot overhead and UE power consumption

A critical feature for UEs with fewer transmit chains than receive antennas. Antenna switching allows the UE to sound all receive antennas by time-multiplexing transmissions across different physical antenna ports.

• 1T2R: One transmit chain, two receive antennas — UE transmits SRS on each antenna in alternating slots
• 1T4R: One transmit chain, four receive antennas — enables full downlink 4-layer MIMO even with a single uplink PA
• 2T4R: Two transmit chains, four receive antennas — common in high-end handsets
• Essential for achieving peak downlink throughput in TDD massive MIMO deployments