Measuring Micro Volt Noise with your Sound Card!

In this video I show how I've been tracking down uV noise in my preamp circuit using my sound card.

[Proudbmodest]

One other thing this video shows is the limitation of the sound card. The noise floor for an Intel spec sound card is 90 dB. This implies a resolution of 10^4.5 = 31, 623 or roughly 15 bits. (90db = voltage ratio of 10^(90/20, log(10^4.5)/log(2) = 14.95).
Even though the sound card ADC is touted as being 24 bits, in reality it's only 15 bits. This adds nuance to the battle between analog and digital. If an analog recording is done with greater than 90 dB dynamic range and with a noise floor below 90 dB, it will be better than a PC sound card can reproduce. But, the 24 bits, although noise limited, gives a huge dynamic range of 144 dB which is difficult to impossible for analog methods.
For measurement purposes, the 15 bit resolution is a hard limit. This is important if the sound card is used for measuring harmonic distortion as it limits the measurement precision and is not suitable for very low THD measurements.
For noise measurements this is not a problem because the signal level can be moved away from the noise floor by using a low noise amplifier to take advantage of the high dynamic range. For noise, the measurement resolution is not that critical.

[Proudbmodest]

The sound card has protection diodes built in, but they have limited current capability. You can take advantage of those diodes by just putting a large resistor in series with the input. This will cause a gain error because the sound card input impedance is around 47k, but this can be compensated for. Also, use a small fast blow fuse in series for further protection.
Another possibility is to use a sacrificial, low noise preamplifier in front of the sound card. Unity gain is sufficient so a low noise bipolar emitter follower circuit would be sufficient or if higher noise resolution is required, add some gain.
Although the sound card dynamic range is 24 bits, it's fundamental resolution is roughly 12 uV peak to peak. All the gain settings are purely digital and have no effect on sensitivity or noise and are just multiplication factors.
For ultra low noise measurement, an ultra low noise preamplifier is required to boost the input above the inherent sound card noise.
The maximum sound card input is about 3V peak to peak which makes it good for a lot of measurement purposes.
The left and right sides of one channel are almost perfectly phase matched which makes it possible to use a sound card for making things like LCR meters with better than 1% precision. Phase, amplitude and frequency can be precisely measured using an FFT.
By using long sampling periods, high frequency resolution is possible. Sampling for one second will give 1 Hz resolution which is handy when measuring the RMS noise level over a given 1 Hz bandwith.
The side gains are not precisely matched, but a simple calibration can be done by using the sound card output as a source and then comparing the measurement results of the two sides and adjusting the gain accordingly. This is handy when doing precision measurements when using the sound card output as a source. One side is dedicated to measuring the source voltage while the other side makes the actual measurement. The sound card uses a bandgap reference so the measurements are quite stable over time.
Also, the sound card has two input channels, each with two sides giving a total of four inputs. The mic inputs have a DC bias voltage applied through a resistor for powering condenser microphones and a series capacitor might be needed if the voltage will cause problems. The bias resistor also lowers the input impedance.
All four inputs are normally not available simultaneously but can be accessed in Linux based PCs using ALSA, but this requires a configuration file in the home directory. This may also be possible in a Windows PC, but with Linux and ALSA available for free, it's not worth the effort.
Check out the o'scopepy project on sourceforge.net This is a Linux and ALSA based sound card oscilloscope with plug in tools like a frequency selective voltmeter and LCR meter. It's a hobby project, but it can be the source for other ideas.

[leiferickson]

Hi Doug! I ran some simulations with 2 x 0.047uF caps and back to back 1N4148's to see what might be done about this. First, let me say that I used ideal parts for the passives for one variety of a worst case setup. I used a 400V Pulse with a 1uS rise and fall time. I found that adding a resistor of 100 or 1k in front of the diodes appears to drop the input voltage presented to the output. Without the resistor there was a 54 volt spike and 90 amps of inrush!. So I think by using this approach and noting the output voltage and taking care to not over current the diodes this might be a sweet accessory! I really love your content! Thanks for posting all of these videos! (Edit - don't use 1N4148's, those were just at the top of LTSpice's list of diodes)

[EdgarsLS]

I made a signal tracer with an output for my PC, it has a max gain of around 600, and input impeadane is above 50Megs because of a CF as the first stage.
and I have 2 back to back leds in paralel with the output for protection. works great!

[leiferickson]

Great video. I fried my laptop soundcard using it as a scope about 20 years ago :-) If you do not have fast rising/falling transients this will probably not kill your soundcard - i.e. the power suply you built comes up to voltage slowly (unless that is averaging on your scope). But if you are connecting the probe to a live power supply you might find that those slow diodes may let through high voltage briefly before they clamp so be carefull. You can consider ESD type of diodes, which may afford a bit more protection. I built a front end preamp for my scope using a 5532 to get down there to see noise. I run it off of batteries. It is really nice using that FFT program on your latop though.

[harveyellis]

Cool idea and project. For audio, it really is better to use linear, rather than SMPS.

[tomekichiyamamoto]

Why did you use an RCA instead of a XLR?

[A_RosnerNZ]

The first power supply you measured, the 300v 5K (I make that 60mA BTW) - what power supply was that? Also, did you try pulling different current from it to see how that affected the noise performance?