TMS320F28069M: Resolver Noise

1. Why you can't simply "lower R6/R7 to cut noise" in the excitation amplifier

• Sets the inverting gain together with R7: A = R7/R6 = 470 kΩ/100 kΩ = ‑4.7
• Sets the input impedance seen by the preceding stage (the analog phase‑splitter that generates the complementary EXC_P/EXC_N sine waves from Q2–Q5)
• Forms the AC‑coupling high‑pass pole with C2: fc1 = 1/(2π·R6·C2) = 160 Hz

1. You load down the phase‑splitter stage. The differential excitation scheme only works correctly if VEXC+ = −VEXC− precisely (so that VR = 2×VEXC at the resolver). Dropping the input impedance of the inverting stage pulls more current from the splitter and can unbalance the amplitude/phase matching between the two excitation channels — which is a bigger accuracy risk than the resistor thermal noise you're trying to remove.
2. You shift fc1 upward unless C2 is scaled up to compensate, risking attenuation/phase shift of the actual 5 kHz excitation tone itself.
3. Any resulting mismatch between the networks matters more than the resistor's own noise. This is the same tradeoff Jiri (the TIDA‑01527 author) flagged on the "Questions to resolver feedback winding monitoring" E2E thread, when a customer proposed lowering resistor values on the feedback (not excitation) side for the same noise‑immunity reasoning. His response there is directly applicable: "capacitors come with relatively high tolerance even if they come from the same reel. Mismatch between the networks ruin the common mode rejection ratio," and a Monte Carlo run with just 1% R / 5% C tolerance showed CMRR collapsing to ~30 dB by 30 kHz. The lesson generalizes: resistor-value changes that aren't carefully re-matched/re-verified tend to buy you worse real-world performance than the noise floor they were meant to fix and can also introduce op‑amp stability/phase‑margin issues (also called out in that same thread for added feedback capacitors).

2. Does excitation noise affect angle repeatability?

• A resolver is ratiometric by design — angle is derived from the ratio of the SIN/COS secondary outputs to the excitation reference, which is exactly why resolvers "suppress common‑mode noise" (per the TIDA‑01527 System Description). Noise that appears identically on both excitation phases, and therefore on both SIN and COS proportionally, largely cancels out in the arctangent angle calculation.
• However, the design's own accuracy data shows that imbalance — not noise amplitude itself — is what directly degrades angle accuracy. Figure 16 in the user guide ("Effect of DC Offset and Gain Imbalance on Angle Error") plots this explicitly:
  • A 1% gain imbalance between the two difference amplifiers introduces a quantifiable non‑linear angle error (Equation 20).
  • A DC offset of VOFFSET = 0.05 × VS(MAX) similarly introduces angle error.
  • This is the mechanism by which excitation noise would translate into angle error: if noise/distortion on EXC+ and EXC− isn't symmetric (e.g., injected asymmetrically through an unbalanced inverting stage, mismatched filter components, or amplifier instability under inductive loading — all of which have shown up in other TIDA‑01527 threads), it behaves like a gain/offset imbalance and directly reduces repeatability.
• For reference, the as‑built TIDA‑01527 (with its stock R6/R7/filter values and matched dual‑amplifier excitation path) achieves ±0.25° angle accuracy, improved to ±0.1° using the design's scattered‑signal‑processing method (Table 1, Key System Specifications) — i.e., noise/imbalance effects are already bounded to that level under nominal, correctly matched conditions.
• The design also explicitly monitors the excitation output via OEXC (ADCINA1), which exists specifically so the firmware can track excitation signal integrity rather than assume it's perfect — reinforcing that excitation fidelity is treated as a real factor in the measurement chain, not something resolvers make irrelevant.