SPORT: Spherical‑PSNR–Optimized tRuncaTion for Power‑Efficient 360‑Degree Video Systems
Md. Sajjad Hossain, Hasibur Rahman Hemel, Kyle Mooney, Yiwen Xu, William Oswald, Mario Renteria-Pinon, Hritom Das, Zhenlin Pei, Jinhui Wang, and Na Gong · 2026
Read SPORT: Spherical‑PSNR–Optimized tRuncaTion for Power‑Efficient 360‑Degree Video Systems by Md. Sajjad Hossain, Hasibur Rahman Hemel, Kyle Mooney, Yiwen Xu, William Oswald, Mario Renteria-Pinon, Hritom Das, Zhenlin Pei, Jinhui Wang, and Na Gong (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-24916.txt.
Introduction
Md. Sajjad Hossain and nine colleagues present SPORT, a framework that cuts the memory power of 360 degree video in virtual reality headsets. Memory bandwidth eats thirty to forty percent of a standalone headset's power. SPORT truncates pixel bits outside the field of view, keeping only the most significant ones. Its optimisation targets the spherical quality metric WS-PSNR directly.
A gaze predictive tile classifier compensates for pipeline latency of about nine milliseconds. It cuts boundary misclassifications by about five points for a tiny time cost. One variant keeps the viewed region lossless and saves about 48 percent of memory power and bandwidth across 4K sequences. The fully adaptive variant reaches about 52 percent savings and beats a standard PSNR based optimiser at equal quality. The design was validated on a fabricated test chip in 130 nanometre CMOS, matching the software model byte for byte.
Catalog record
| CREATOR | Md. Sajjad Hossain, Hasibur Rahman Hemel, Kyle Mooney, Yiwen Xu, William Oswald, Mario Renteria-Pinon, Hritom Das, Zhenlin Pei, Jinhui Wang, and Na Gong |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.24916
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
SPORT: Spherical‑PSNR–Optimized tRuncaTion for Power‑Efficient 360‑Degree Video Systems Md. Sajjad Hossain, Hasibur Rahman Hemel, Kyle Mooney, Yiwen Xu, William Oswald, Mario Renteria-Pinon, Hritom Das, Zhenlin Pei, Jinhui Wang, and Na Gong Abstract Memory bandwidth accounts for 30–40% of total power consumption in standalone virtual reality (VR) headsets, yet existing systems typically store the entire 360 ∘ frame at a uniform resolution regardless of viewer gaze. This paper presents SPORT ( S pherical- P SNR O ptimized t R unca T ion), a bit-truncation framework that reduces display-path memory power by storing only the most significant bits of pixels outside the user’s field of view (FoV). Specifically, a new bit-truncation framework is developed to use weighted-to-spherically-uniform PSNR (WS-PSNR) directly in the optimization constraint, eliminating the metric inconsistency that arises when standard PSNR is used for a WS-PSNR quality target. Also, gaze-predictive tile classification compensates for the 9.33 ms end-to-end pipeline latency, reducing boundary misclassifications by 5.2 percentage points at a cost of only 0.01 ms. In addition, the developed SPORT-B variant which keeps the FoV lossless achieves 47.9% memory power saving and 47.9% bandwidth reduction across different 4K video sequences while satisfying all three per-region WS-PSNR thresholds and maintaining SSIM = 1.000 in the attended region. The full adaptive variant SPORT-A reaches 51.6% power saving, 3.1 percentage points more than a PSNR-based optimizer at equal measured quality. SPORT is validated on the TrunMEM360 flexible SRAM Application-Specific Integrated Circuit (ASIC) fabricated in SkyWater 130 nm CMOS, confirming byte-exact silicon–software agreement, with WS-PSNR and SSIM matching within 0.1 dB and 0.001. CACTI-based analysis confirms 48.72% DRAM leakage reduction and 36.4%/36.7% read/write energy reduction. The total motion-to-photon latency of 9.33 ms satisfies the 20 ms VR comfort budget with a 53.3% safety margin. I Introduction Standalone virtual reality headsets run on batteries to render 360° video at 90 frames per second (fps) while maintaining a total motion-to-photon latency under 20 ms to avoid user discomfort [ 1 ] . This imposes significant pressure on the memory subsystem to simultaneously meet both requirements. For example, a single frame of 4K equirectangular projection (ERP) video takes about 22.1 MB per frame and the display-path memory must support about 2 GB/s of read and write activity at 90 fps. The memory subsystem alone uses 30-40% of the headset’s total power [ 2 ] , which directly limits the current devices’ battery life to two to three hours. This work is motivated by a crucial observation: at any given time, the viewer’s foveal attention is limited to a cone of about 45° around the gaze direction. The remaining 80% of the frame lies in regions where fine pixel-level detail is imperceptible [ 1 , 3 ] . However, existing system typically stores the full frame with uniform resolution (e.g., 8-bit), incurring excessive bandwidth and power consumption regardless of the perceptual relevance of different regions of the frame. One promising solution to this problem is bit truncation at the memory write stage, where the least significant bits (LSBs) of pixels in peripheral regions are discarded before writing to the off-chip memory (DRAM), and an approximate value is reconstructed during readback. For example, Kim and Kyung [ 4 ] established the theoretical foundation for 2D video, showing that replacing truncated bits with the binary pattern 10…0 minimizes the expected reconstruction error. Although this technique has been applied in prior work, e.g., [ 5 , 6 ] , three fundamental limitations prevent its application for 360° videos. First, existing bit-truncation methods optimize using standard PSNR, while 360° video is evaluated using WS-PSNR [ 7 ] . This is responsible for ERP distortion by weighting each pixel by the cosine of its latitude. The same numerical threshold (e.g., 40 dB) can cause different perceptual implications under the two metrics. This mismatch leads to suboptimal truncation decisions where the optimizer meets its PSNR target but fails the WS-PSNR requirement [ 8 ] . This metric inconsistency is particularly problematic for 360° videos, as polar regions are over-sampled and equatorial regions where viewers focus are under-sampled. Second, the end-to-end motion-to-photon latency of approximately 9.33 ms implies that by the time a decoded frame is written to DRAM and displayed by the time the viewer has already moved their head. Tile classification against the head movement prediction can mitigate the resulting misclassifications by adjusting the quality of boundary regions during head movement. Third, prior truncation memories are designed for fixed truncation levels without runtime adaptability. As a result, they cannot dynamically adjust truncation per region based on real-time gaze prediction or content-dependent quality requirements. This forces a trade-off between aggressive truncation that degrades FoV quality and conservative truncation that sacrifices potential power savings. In addition, existing studies typically lack silicon-based hardware testing and validation. This paper addresses all of the above limitations through the SPORT framework using a software-hardware co-design. The key contributions are summarized as follows. • WS-PSNR-consistent truncation framework: To eliminate the metric inconsistency, the current bit-truncation theory is extended to directly use WS-PSNR in the optimization constraint (Section IV ). It is demonstrated that the ideal 10…0 dummy pattern does not require hardware modification under latitude weighting (Theorem 2 ). • Gaze-predictive region segmentation: Boundary misclassifications are reduced from 8.3% to 3.1% with only 0.01 ms of added cost by classifying tiles using head movement prediction (Section V ) to account for the 9.33 ms pipeline latency. • TrunMEM360 heterogeneous memory architecture: A heterogeneous three-bank SRAM architecture is proposed (Section VI ) with per-bank truncation managers and Head/Tail power-gating to eliminate the leakage and switching power for truncated bits to handle three regions of 360° videos separately. • Fabricated ASIC validation: The TrunMEM chip (130 nm CMOS) validates the SPORT framework on the silicon level with WS-PSNR and SSIM matching software simulation within 0.1 dB and 0.001 across all test frame tiles (Section VIII ). • Comprehensive 4K simulation: Software simulation on real head-tracking traces (Section VII ) validates SPORT-B with 47.9% power saving and 47.9% bandwidth reduction while maintaining SSIM = 1.000 in FoV for real VR application. • CACTI DRAM power validation: CACTI-based analysis shows \sim 48.7% leakage reduction and 36.4% read energy reduction when effective memory footprint is halved, validating the proposed analytical power model. • Zero-latency pipeline integration: TrunMEM360 adds zero net latency to the critical path, achieving 9.33 ms total motion-to-photon latency with 53.3% safety margin. The organization of the paper is as follows. Section II reviews related work on 360° video such as quality metrics, viewport-adaptive streaming, memory-level bit truncation, and head movement prediction. Section III describes the proposed SPORT scheme, Sections IV and V develop the mathematical framework for bit truncation and region segmentation, and Section VI describes the proposed TrunMEM360 memory architecture. We discuss the software-level evaluation results in Section VII; Section VIII presents the developed hardware testing platform and discusses its results; and finally, Section IX concludes the paper. II Related Work II-A Quality Metrics for 360° Video Standard PSNR assigns equal importance to all pixels, which is not suitable for ERP content. For example, a pixel at 80° latitude represents approximately 17% of the spherical surface area of an equatorial pixel, yet PSNR treats them equally. Sun et al. [ 7 ]