Resolution Adaptive Quantization for Efficient Signal Processing

19 concepts10 papers2 proposals

Traditional digital signal processing systems—whether processing high-frequency wireless signals in massive MIMO base stations or executing matrix multiplications in compute-in-memory (CIM) accelerators—face a fundamental hardware bottleneck: the analog-to-digital converter (ADC). ADCs bridge the continuous physical world and discrete digital computation, but their power consumption scales exponentially with bit precision ($2^B$).

Historically, engineers designed these systems for the worst-case scenario, provisioning fixed, high-resolution ADCs to handle peak signal-to-noise ratios (SNR) or maximum signal dynamic ranges. This approach wastes massive amounts of energy when conditions are favorable. Resolution-adaptive quantization breaks this rigid paradigm by dynamically tuning the precision or the quantization thresholds of data converters in real time based on the immediate demands of the signal environment.

The Intuition of Dynamic Bit-Precision Scaling

Why use 10 bits of precision when 4 bits will suffice? In massive multi-user MIMO (MU-MIMO) millimeter-wave (mmWave) systems, base stations feature large antenna arrays. When the channel conditions are excellent or when only a few users are actively transmitting, the spatial multiplexing gain provides inherent structural redundancy. As demonstrated by Oscar Castañeda et al. in A Resolution-Adaptive 8 mm2 9.98 Gb/s 39.7 pJ/b 32-Antenna All-Digital Spatial Equalizer for mmWave Massive MU-MIMO in 65nm CMOS, reducing the bit precision of both the ADCs and the subsequent digital spatial equalization engines during these favorable periods yields massive energy savings without degrading communication performance.

By moving away from static configurations, systems can allocate quantization bits across radio frequency (RF) chains based on the instantaneous channel state. For instance, Yue Xiu et al. in Uplink Achievable Rate Maximization for Reconfigurable Intelligent Surface Aided Millimeter Wave Systems with Resolution-Adaptive ADCs and Hualian Sheng et al. in Energy Efficiency Optimization for Millimeter Wave System With Resolution-Adaptive ADCs show that dynamically matching ADC resolution to real-time channel conditions mitigates quantization noise precisely where needed while slashing the exponential power footprint of the receiver array.

This concept extends beyond wireless communication into hardware accelerators. In analog compute-in-memory architectures, Haobo Liu et al. in AFPR-CIM: An Analog-Domain Floating-Point RRAM -based Compute- In- Memory Architecture with Dynamic Range Adaptive FP-ADC introduced an adaptive floating-point ADC that dynamically shifts its range to handle the wide and unpredictable variations of analog neural network computations, preventing precision loss without incurring the power overhead of a massive fixed-range converter.

Fixed Quantization (Worst-Case Design):
Signal Range:  |-----------------------------------|
ADC Steps:     ||||||||||||||||||||||||||||||||||||| (High power always)

Resolution-Adaptive Quantization:
Low Demand:    |---------|                           (Fewer steps, low power)
High Demand:   |-----------------------------------| (Full steps, high power)

Hardware Mechanisms for Runtime Reconfiguration

To turn this theory into energy savings, the underlying hardware must physically alter its precision on the fly. In a typical Successive Approximation Register (SAR) ADC, the conversion process determines bits sequentially by comparing the input voltage against a reference using a digital-to-analog converter (DAC) capacitor network. To save energy when full precision is unnecessary, Yunfeng Hu et al. in A 9-10-Bit Adjustable and Energy-Efficient Switching Scheme for Successive Approximation Register Analog-to-Digital Converter with One Least Significant Bit Common-Mode Voltage Variation developed a scheme that dynamically toggles between 9-bit and 10-bit modes by selectively disabling specific logic units and their associated capacitors.

Similarly, in mixed-signal processing-in-memory architectures that accelerate deep learning models, different memory cells may demand different processing fidelities. Chang Eun Song et al. in Hybrid SLC-MLC RRAM Mixed-Signal Processing-in-Memory Architecture for Transformer Acceleration via Gradient Redistribution leverage reconfigurable ADCs that support multiple precision levels by explicitly bypassing specific bit comparisons. When processing lower-precision memory arrays, the hardware skips the fine-grained comparison steps entirely, directly translating hardware inactivity into power savings. This tight hardware-software co-design allows an all-digital spatial equalizer, as proposed by Oscar Castañeda et al. in Resolution-Adaptive All-Digital Spatial Equalization for mmWave Massive MU-MIMO, to scale down its power consumption by up to 22× by adapting to real-time network conditions rather than over-engineering for the worst case.

Extracting High Resolution from One-Bit Quantizers

The most extreme manifestation of low-power quantization is the 1-bit ADC, which acts as a simple comparator outputting a single binary value (0 or 1). While 1-bit ADCs drastically simplify hardware and minimize power, they traditionally suffer from severe rate loss and quantization noise, particularly in high-SNR regimes where fine-grained signal details are lost.

Resolution-adaptive quantization solves this through adaptive thresholding. Instead of keeping the zero-crossing threshold static, an adaptive threshold receiver dynamically shifts the comparison threshold based on prior signal observations.

Static 1-Bit Threshold:
Signal:    /\    /\        --> Quantized to: 1, 1, 0, 0
Threshold: ---------- (0V)
           \/    \/

Adaptive 1-Bit Threshold:
Signal:    /\    /\        --> Threshold tracks the signal,
Thresh 2:    ----          --> extracting more information
Thresh 1:  --    --        --> over successive time steps.

In a foundational work, Abbas Khalili et al. in On Multiterminal Communication over MIMO Channels with One-bit ADCs at the Receivers introduced a receiver architecture that utilizes analog delay elements and a strictly upper-triangular coefficient matrix to perform temporal signal processing before quantization. By adjusting the 1-bit ADC thresholds based on previously received signal instances, the receiver effectively tracks the signal's trajectory over time.

This temporal feedback loop fundamentally alters the power-rate trade-off. As shown by Abbas Khalili et al. in On Throughput of Millimeter Wave MIMO Systems with Low Resolution ADCs and further detailed in MIMO Networks With One-Bit ADCs: Receiver Design and Communication Strategies, this architecture allows a nominal 1-bit converter to achieve a significantly higher effective resolution. By extracting more information across successive channel uses, the adaptive threshold receiver completely eliminates the high-SNR rate loss typically associated with coarse quantization, achieving near-optimal data rates irrespective of the channel's rank.

Key Foundational Work

On Throughput of Millimeter Wave MIMO Systems with Low Resolution ADCs (2020) — contributed 3 concept elements to this direction.
On Multiterminal Communication over MIMO Channels with One-bit ADCs at the Receivers (2019) — contributed 3 concept elements to this direction.
Energy Efficiency Optimization for Millimeter Wave System With Resolution-Adaptive ADCs (2020) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_resolution_adaptive_quantization_cf1639d4,
  title = {resolution-adaptive quantization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/resolution-adaptive-quantization/rdr_cf1639d4057e47ae87d39c26781f0bf4},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_resolution_adaptive_quantization_cf1639d4_v1,
  title = {resolution-adaptive quantization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/resolution-adaptive-quantization/rdr_cf1639d4057e47ae87d39c26781f0bf4/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake