Customer evidence

Use these recorded scorecards to shortlist codec configurations, then evaluate them on representative signals from your application. Compression ratio (CR) describes payload reduction; it does not describe reconstruction quality by itself.

What these results cover

These tables are recorded evaluation snapshots, not a live inventory of downloadable models. Sample counts and metric definitions must be read alongside the values. Refreshing the site does not run an evaluation.

• PPG: 5-second frames at 64 Hz. Sample counts vary by ratio.
• ECG: 2-second frames at 256 Hz.
• Quality: this page uses noise-aware scorecards. The model pages also show a separate validation and long-recording evaluation; those values should not be treated as the same run.
• Package size: encoder, decoder, and codebook files. These sizes do not measure runtime RAM, firmware size, or inference latency.

Compression and quality

Lower error is better within the same evaluation. N is the reported sample count. A dash means the measurement was not reported, not zero.

PPG quality

Data source: measurements preserved from the original documentation.

CR	N	Truth PRD (%)	Faithful PRD (%)	HR error (bpm)
2x	9788	0.62	2.59	0.19
4x	9789	1.14	3.44	0.29
8x	9789	2.66	5.10	0.49
16x	9789	5.75	7.54	0.88
32x	6336	12.35	12.08	1.79

ECG quality

CR	N	Truth PRD (%)	Faithful PRD (%)	HR error (bpm)
2x	1000	1.38	2.31	0.48
4x	1000	2.33	3.63	0.95
8x	1000	3.84	6.03	1.39
16x	1000	7.28	9.72	2.05
32x	1000	13.27	14.59	2.34
64x	1000	20.49	21.05	3.00

Package footprint

The reported total excludes validation samples and documentation. Component sizes are rounded independently, so their displayed sum may differ from the total.

PPG files

CR	Total	Encoder	Decoder	Codebook
2x	88 KiB	5 KiB	19 KiB	65 KiB
4x	147 KiB	28 KiB	54 KiB	65 KiB
8x	196 KiB	39 KiB	92 KiB	65 KiB
16x	300 KiB	54 KiB	180 KiB	65 KiB
32x	267 KiB	54 KiB	180 KiB	33 KiB

ECG files

CR	Total	Encoder	Decoder	Codebook
2x	88 KiB	5 KiB	19 KiB	65 KiB
4x	147 KiB	28 KiB	54 KiB	65 KiB
8x	196 KiB	39 KiB	92 KiB	65 KiB
16x	300 KiB	54 KiB	180 KiB	65 KiB
32x	267 KiB	54 KiB	180 KiB	33 KiB
64x	397 KiB	76 KiB	289 KiB	33 KiB

Detailed signal measurements

These metrics help investigate differences that a single waveform score can miss. Definitions and measurement limitations are in the validation scorecard.

PPG detail

CR	PRDN-noise (%)	Band error	Coherence	Seam ratio
2x	0.15	0.0155	0.9928	-
4x	0.23	0.5885	0.9861	-
8x	0.64	0.6623	0.9565	-
16x	0.79	12.5676	0.8991	-
32x	0.20	0.0356	0.8014	-

ECG detail

CR	PRDN-noise (%)	Band error	Coherence	Seam ratio
2x	0.00	0.0081	1.0000	-
4x	0.00	0.0104	1.0000	-
8x	0.00	0.0144	1.0000	-
16x	0.04	0.0246	1.0000	-
32x	0.78	0.0497	1.0000	0.176
64x	3.53	0.0835	1.0000	-

Choose a candidate

1. Shortlist ratios that meet your payload budget.
2. Inspect the waveform and heart-rate errors together. Check sample counts and noise conditions before comparing rows.
3. Examine the PPG or ECG plots and the detailed PPG or ECG noise buckets.
4. Run the evaluation notebook on your recordings, then validate a deploy package.

No fixed ratio is recommended for a clinical or product use case by these aggregate measurements alone.