Pt. 2: Analogue Improvements

Pt. 2: Analogue Improvements

Introduction

As described in the first blog post, v2 began as an effort to improve manufacturability and future‑proof the design. As those updates grew in scope, I also reworked the analogue section to achieve further gains in performance. 

The improvements and data presented here come from tests on three v1 units and the three first v2 prototypes made, so none of them were specially selected in any way.
The v1 group includes an early serial number v1.0 unit, the final v1.4 unit from Ian I received when I took the design over, and the first v1.6 unit I built with the initial wrytech design changes (v1.5 was an unproduced intermediary). 
On the v2 side, the first three production units were measured to confirm performance before moving forward with production. 

All tests were conducted with each unit warmed up, powered by charged batteries, and set to  10 V output.

Summary of Improvements

	Tempco (ppm/K)	0.1–10 Hz Noise (nV RMS)	10 Hz–10 kHz Noise (µV RMS)
Average v2	–0.1172	415.4	4.611
Average v1	–0.2689	809.4	12.17
Improvement	56.4 %	48.7 %	62.1 %

Optimized Step Response

v1 units typically overshoot ~25 % on a voltage change step and settle in ~10 ms. v2 eliminates overshoot and achieves a ~3 ms rise time. Pictures show a 0 to 1 V steps as example.

Low‑Frequency Noise (0.1–10 Hz)

Measured with a DIY LNA and oscilloscope over 2 minutes; average RMS values taken over. The included measurement with the 50 Ω short confirms the noise floor is well below the readings taken.

Measurement	v2 Smpl 1	v2 Smpl 2	v2 Smpl 3	v1.0	v1.4	v1.6
0.1–10 Hz Noise (nV RMS)	460.1	402.4	383.7	796.6	827.8	803.8

Wideband Noise (10 Hz–10 kHz)

Measured on a QuantAsylum QA403 at 0 dBV FS and 48 kHz sample rate for minimized noise floor; Measurement with 50 Ω short shows noise floor.

Measurement	v2 Smpl 1	v2 Smpl 2	v2 Smpl 3	v1.0	v1.4	v1.6
10 Hz–10 kHz Noise (µV RMS)	4.794	4.644	4.395	9.735	13.460	13.300

Temperature Coefficient

Early Tempco measurements (–0.23 ppm/K) by xDevs have now been confirmed across three v1 units, with v2 units showing further improvement.
Tests were run from 20 °C to 50 °C in a DIY thermal chamber (Peltier + forced air), using an Arroyo 5400 TEC controller and a Keithley 2002 meter. After a 30 min soak at 20 °C, temperature was ramped at 2 °C/min to 50 °C and held for 1 h.
The graphs below plot raw voltage and a 100‑point rolling average. The endpoints used for the coefficient calculation are highlighted and span a 30K temperature difference.

Measurements of the v1 units were done first, with some changes in the script for the v2 units to achieve a better temperature curve. Tuning the control loop is somewhat of a challenge, as response time is slow.
Some noncritical data is missing in the 50 °C plateau phase for the v1.4 and v1.6 units due to an error in the script, resulting in straight lines connecting the last and first valid data points.

Measurement	v2 Smpl 1	v2 Smpl 2	v2 Smpl 3	v1.0	v1.4	v1.6
Tempco (ppm/K)	–0.0703	–0.0917	–0.1305	–0.2800	–0.2933	–0.2333

Summary

Although the sample size is modest, the improvements are clear and the results consistent within both groups of units: v2 units exhibit a tighter temperature coefficient, lower low‑frequency and wideband noise, and a cleaner step response.
None of this required a radical circuit overhaul - rather, it’s the result of careful board‑level refinements across successive v2 prototype layouts, all while preserving the proven core topology.
In addition, initial temperature coefficient measurements for v1 prototype units could be confirmed, rounding off the characterization of these (now‑legacy) units.