Electronic – TS317 voltage regulator falling voltage

Observations from the data:

• First row of data appears to be a statistical outlier and can be rejected: All readings are lower than the profile for subsequent rows, probably due to start-up behavior.
• The last band is not saturating, it varies from 1002 to 1022, so it is capturing valid data
• Bands 4, 5 and 6 are nearly saturated, but are not "stuck to the rail", they do show some variation, so they aren't in a failed, stuck-to-Vcc state

This indicates that the signal generated by the preamplified electret microphone is very high at the frequency bands represented by output 4, 5 and 6. This can be verified using a reverse biased avalanche diode or a zener diode (V_zener > 7 Volts) as a white noise source replacing the electret microphone in the BoB's preamplifier, and checking the output using an oscilloscope in spectrum analysis mode.

This is not surprising at all, as electret microphones are not very linear, and the preamplifier in the break-out board does not have any equalization built in, as per SparkFun's schematic. Hence net result is expected to be a rather non-linear output profile.

The MSGEQ7 frequency response, on the other hand, is pretty linear across bands:

If the problem just involves getting the signal levels for all bands into usable range, a simple resistor voltage divider to attenuate the incoming signal to perhaps 50 or 75% should do the trick, across all bands. 50% will leave plenty of headroom to capture signal spikes on any band, as will invariably occur with normal audio.

^{simulate this circuit – Schematic created using CircuitLab}

If it is also necessary to equalize the signal such that the mid-range and tweeter bands are attenuated while the relatively weak woofer bands are not affected, then a circuit such as a bandstop filter can be used, with a very low Q, and the pole frequency between the band-5 (~ 2200 Hz) and band-6 (~ 6000 Hz) frequencies. In effect, this will be more of a "band attenuate" filter, for the problematic bands, with some effect on the lower and higher bands as well:

_(source)

Such a filter is non-trivial, and perhaps deserves a separate question if required.

You have noise in your system, plain and simple. How you handle it is the same ways that you handle any other noise: remove the source of the noise, or filter out the noise. Some things to consider are:

• Long cables can cause noise. Shorten them, or use shielded cables
• Use differential measurements, if possible.
• Make sure that you use good power distribution, routing, and filtering.
• Make sure that your PCB is designed to reduce noise on the important signals.
• Placing a cap on your ADC input pin, between the pin and GND, can help reduce noise from a pot. This C, plus the R that is your pot, form an RC low pass filter. The larger the cap, the better the filtering (but the slower the system will respond to changes of the input). I would start with a 0.1 uF cap, and go up from there.
• Use a lower-value POT. 10K, 5K, or even 1K ohm. This effectively decreases the output impedance of your sensor (the pot) and reduces the chances of picking up noise. Make sure you do not exceed the power limits on the pot.
• Filter the signal in software. It could be as simple as having a moving average, or you could do a more complex form of filter. The longer time that you average the signal, the better effective resolution you will get, with the obvious down-sides.
• Make sure that you are operating the ADC in the MCU correctly.