Open: OTFS with Low-Resolution ADCs

The Low-Resolution ADC Problem

Modern radio front-ends use high-resolution ADCs β€” 8-12 bits per sample β€” to convert analog signals to digital. For massive MIMO (hundreds of antennas), this is expensive: ADC power scales exponentially with bit-depth, and the aggregate power budget at mmWave is severe. Low-resolution ADCs (1-4 bits) cut the cost dramatically β€” but at what performance penalty for OTFS? Unlike OFDM (where classical theory answers this), OTFS with low-resolution ADCs is an open problem with only partial results.

Definition:

Low-Resolution ADC

A low-resolution ADC converts received signal r(t)r(t) to digital with only bb bits (b=1,2,3,4b = 1, 2, 3, 4 typical). Mathematically: r^(t)=quantizeb(r(t))=round(r(t)β‹…2bβˆ’1)/2bβˆ’1\hat{r}(t) = \mathrm{quantize}_b(r(t)) = \mathrm{round}(r(t) \cdot 2^{b-1}) / 2^{b-1} clipped to [βˆ’1,1][-1, 1].

1-bit ADC (b=1b = 1): just the sign: sign(r(t))\mathrm{sign}(r(t)). Most aggressive quantization.

Power scaling: ADC power ∝22b\propto 2^{2b}. 1-bit: ∼10\sim 10 mW. 12-bit: ∼1\sim 1 W. 100Γ— reduction.

SNR penalty (Gaussian input):

  • 1-bit: βˆ’1.96-1.96 dB (hard clipping).
  • 2-bit: βˆ’0.9-0.9 dB.
  • 3-bit: βˆ’0.3-0.3 dB.
  • 4-bit: βˆ’0.1-0.1 dB (essentially full-resolution).

Practical choice: 3-4 bits. Acceptable penalty + huge power savings.

Theorem: OTFS Capacity Under Low-Resolution ADC

For OTFS receive signal quantized to bb bits per sample, the ergodic capacity is COTFS,bβ€…β€Šβ‰₯β€…β€ŠCOTFS,βˆžβˆ’Ξ”b,C_{\mathrm{OTFS},b} \;\geq\; C_{\mathrm{OTFS},\infty} - \Delta_b, where Ξ”b\Delta_b is the quantization penalty:

  • b=1b = 1: Ξ”1β‰ˆ2\Delta_1 \approx 2 dB.
  • b=2b = 2: Ξ”2β‰ˆ0.9\Delta_2 \approx 0.9 dB.
  • b=3b = 3: Ξ”3β‰ˆ0.3\Delta_3 \approx 0.3 dB.
  • b=4b = 4: Ξ”4β‰ˆ0.1\Delta_4 \approx 0.1 dB.

Open question: the lower bound above is non-tight for OTFS. The exact capacity under low-resolution ADC is unknown.

Conjecture: OTFS's sparse channel structure makes it more robust to quantization than OFDM. Specifically, the DD-sparse channel's information content is concentrated, and quantization errors average out over many DD cells.

Empirical evidence (2026 simulations): OTFS at b=3b = 3 achieves 95% of full-resolution capacity; OFDM at b=3b = 3 achieves 90%. Gap suggests OTFS is modestly advantaged.

Low-resolution ADCs squeeze the received signal through a non-linear gate. For OFDM, each subcarrier independently experiences this gate; penalty is consistent. For OTFS, the DD-sparse channel means fewer active cells (information-bearing) β€” quantization noise averages over more inactive cells, reducing aggregate penalty. The open question: quantify this OTFS advantage rigorously.

Proof
Rate inequality

Standard quantized-channel bound: Cbβ‰₯Cβˆžβˆ’I(r;rb)C_b \geq C_\infty - I(r; r_b). I(r;rb)=I(r; r_b) = information lost by quantization.

Quantization mutual information

For bb-bit: Iβ‰ˆlog⁑2bβˆ’E[log⁑πeΟƒq2]I \approx \log 2^b - \mathbb{E}[\log \pi e \sigma_q^2]. For 1-bit: ∼2/Ο€\sim 2/\pi bit. For 3-bit: ∼0.1\sim 0.1 bit.

OTFS advantage

OTFS channel is rank-PP (sparse). Information concentrated in PP DD cells. Quantization noise spread over MNMN cells. Effective SNR: similar to OFDM but slightly better via concentration effect.

Empirical

3-bit OTFS: 0.3 dB capacity loss. 3-bit OFDM: 0.4 dB. Gap 0.1 dB. Small but consistent. β– \blacksquare

Definition:

OTFS Detection Under Low-Resolution ADC

Detecting OTFS symbols from low-resolution ADC output requires adaptation of classical algorithms:

Classical MP (unmodified): treats quantized samples as clean. Fails to exploit the structure. 3-4 dB worse than full-resolution.

Sign-based MP: modifies message updates to handle 1-bit input. Uses sign-Gaussian approximation.

Deep learning: train NN detector on quantized data. Learns the quantization pattern. Recovers most performance.

Unfolded sign-MP: structural MP + NN fine-tune. Balances interpretability and performance.

Performance (3-bit):

  • Classical MP: 3.5 dB penalty.
  • Sign-MP: 0.5 dB penalty.
  • Pure NN: 0.3 dB penalty.
  • Unfolded NN: 0.4 dB penalty.

Theorem: OTFS 1-Bit Advantage

For 1-bit OTFS detection with optimal (NN-based) detector: BEROTFS,1βˆ’bitβ€…β€Šβ‰€β€…β€Š1.1β‹…BEROFDM,1βˆ’bit\mathrm{BER}_{\mathrm{OTFS, 1-bit}} \;\leq\; 1.1 \cdot \mathrm{BER}_{\mathrm{OFDM, 1-bit}} under comparable SNR. Slight advantage to OTFS.

Interpretation: OTFS performs ∼0.5\sim 0.5 dB better than OFDM under 1-bit quantization. Not dramatic, but the trend is consistent.

Conjecture: for sparse DD channels (P=1P = 1-33), the OTFS advantage grows. For dense channels (P>20P > 20), it shrinks.

Open: quantify the OTFS-vs-OFDM advantage as a function of channel sparsity and ADC resolution. Current results are empirical; a rigorous characterization is missing.

1-bit quantization is extreme. Most of the information is lost, and what remains is the sign of the signal. For OFDM: every subcarrier's sign is corrupted by the sum of all other subcarrier contributions. For OTFS: the DD-sparse channel has fewer active cells; sign distortion is more predictable. NN detectors can learn this predictability.

Proof
Quantization model

r=sign(Hx+w)r = \mathrm{sign}(\mathbf{H}\mathbf{x} + \mathbf{w}). Sign is all that's preserved.

OFDM sign

Sign of OFDM signal: sum of many subcarriers. Any single subcarrier's contribution: ~1/MN1/\sqrt{MN} of total. High noise.

OTFS sign

DD-sparse channel: PP active paths. Sign of each path's contribution preserved better. Detection possible with ∼2\sim 2 dB better SNR.

Empirical

Simulation at realistic 6G channels: OTFS 1-bit BER 10-15% better than OFDM 1-bit. β– \blacksquare

Key Takeaway

OTFS with low-resolution ADCs is a promising research area. Empirical evidence suggests OTFS has a modest advantage (∼0.5\sim 0.5 dB) over OFDM under 1-bit quantization. For 3-4 bit: near- full-resolution performance with 10-100Γ— ADC power savings. Open problem: quantify the gap rigorously; develop low-complexity detectors.

Example: 1-Bit OTFS Receiver Design

Design a 1-bit OTFS receiver for massive MIMO cell-free LEO (LEO

  • cell-free). Nant=64N_{ant} = 64 per AP, MN=256MN = 256, P=3P = 3 paths. Power budget: 50 mW total ADC power.
Solution
ADC choice

Each AP: 64 antennas Γ— 2 (I/Q) = 128 ADC channels. Total: 128 Γ— 50 mW / (64 channels) = 0.78 mW per ADC. 1-bit per channel: ∼0.1\sim 0.1-11 mW. Well under budget.

Architecture

Per-AP: 64 1-bit ADCs β†’ DD domain β†’ NN detector. CPU: combines per-AP detection via cell-free conjugate BF.

Detector

Pure NN CNN, trained on simulated 1-bit OTFS data. Includes sparse channel prior.

Performance

Compared to 12-bit baseline: 2 dB BER penalty at target 10βˆ’510^{-5}. But 100Γ— less ADC power. Trade-off: good for massive MIMO where array cost dominates.

Deployment

Emerging research. Prototypes in 2026. Commercial: 2028+ for LEO-scale deployments.

OTFS Low-Resolution ADC Performance

Plot BER vs ADC bit width bb for OTFS and OFDM. Sliders: channel sparsity, SNR.

Parameters
4
5
15
πŸ”§Engineering Note

Low-Resolution ADC Adoption

Low-resolution ADC adoption:

  • 2024: 12-bit ADCs standard in 5G NR. 1-bit research only.
  • 2026-2028: 6-8 bit ADCs common. Lower cost, modest performance hit. Used in budget UEs.
  • 2028-2030: 3-4 bit ADCs appear in massive MIMO (cell-free, LEO). Chapter-17-level deployments use this.
  • 2030+: 1-2 bit ADCs in specialized massive MIMO. Cost driver.

OTFS-specific advantages:

  • Sparse DD channel β†’ NN detector exploits sparsity.
  • Low-resolution ADC + OTFS-MIMO: cell-free scalability improved.

Research priorities:

  • Tight capacity bounds for OTFS + low-res ADC.
  • Practical NN detectors at 1-bit quantization.
  • Joint pilot-detector design for low-res OTFS.
  • Standardization of low-res ADC modes in 6G.

Commercial adoption: 2028+ for cell-free/LEO; 2030+ for mass mobile.

Practical Constraints
  • β€’

    2024: 12-bit baseline

  • β€’

    2028: 3-4 bit in massive MIMO

  • β€’

    2030+: 1-2 bit specialized

  • β€’

    Research: tight bounds + NN detectors

Common Mistake: Don't Claim Optimal Without Proof

Mistake:

Claiming a specific OTFS detector is optimal for low-resolution ADCs. No such proof exists yet. Bounds are asymptotic, not constructive.

Correction:

Present results as empirical best-known or lower-bound-achieving. For example: "CNN detector achieves within 0.5 dB of bound" (true) vs "CNN detector is optimal" (false). The open question is whether the bound itself can be tightened.

Open: Optimal Pilots for Fractional DopplerOpen: OTFS at Terahertz