The QA401 relies on USB power. This is a great convenience because it means setup and making a measurement is quick and easy. But not all USB power is created equal, and it makes sense to understand how USB power quality can impact your measurements. We recently came across an particularly nasty USB supply on a machine in our office.
How Power Supply Noise can Creep Into your Measurements
The QA401 has two independent supply sections, known as the digital and analog supplies. The digital supplies use LDOs to generate 3.3V, 2.5V and 1.25V. The 3.3V is the logic IO voltage, and the 2.5V and 1.25V supplies run the FPGA internal cores. These supplies share the ground with the USB.
The analog supplies do NOT share a ground with the USB. Isolation is achieved by a TI SN6501 push/pull open-loop DCDC converter and a small transformer. Four of these circuits are used to generate +/-6.5V split rail, a 5V section and a 3.3V section. In all cases, the output of the transformer is then fed into an LDO. You can see the QA401 split rail generation in the schematic below.
The regulators used are TPS76301. These are adjustable LDOs from TI, rated for 150mA. In the QA401, there's about 65 mA being pulled from each rail.
The circuit above runs at a switching frequency of about 400 KHz. And the transformers are designed to really like this frequency. If a noise frequency is much lower than the 400 KHz, then the transformer doesn't do a good job of passing it across the gap. And if the frequency is much higher than 400 KHz, a similar situation occurs. But noise on the VUSB bus that is around 400 KHz will make it across the magnetic gap.
Now, the LDOs would ideally take the 400 KHz noise and hammer that down to nothing. There is a spec, called the PSRR, or power supply rejection ratio, that specifies how much noise on a supply line will be suppressed in the output of the LDO. On these particular LDOs, the curve is shown below and we can see that at 400 KHz we might be looking at about 40 dB of rejection. Pretty good, but not great. There are certainly LDOs that can do much better.
But the ADC shouldn't care about noise at 400 KHz, right?
Noise in this region is well above our sample rate, and the ADC has input filters (on-chip and off-chip) to hammer the noise down. But it's still there, although diminished. That noise will then be aliased back into the audio range in the measurement.
And that is how really bad USB noise can degrade a measurement.
OK, so how bad can it be?
As an experiment, we took a QA401 and powered it from a noisy USB port with the inputs shorted. That plot is below. Note the noise floor is about -114.7 dB, and there's clear bump around the 1 KHz region. Remember, we don't know if the bump at 1 KHz means there's USB noise at 1 KHz OR if it means there's noise at a higher frequency that is being aliased down to 1 KHz. It's most likely the latter, because 1 KHz noise would have a tough time getting across the magnetic gap and the PSRR at 1 KHz is pretty good.
March 14 Edit: Take a look at the peak amplitude: It's about 655 nV. If the USB supply noise entered the QA401 at 50 mVp (estimated via quick scope measurement), and was knocked down to 655 nV, then that means that there was about 94 dB of suppression. The LDOs were optimistically doing 50 to 60 dB via PSRR, which suggests another 30 to 40 dB was coming from post-rectifier caps in the DCDC.
Repeating the measurement with a lab supply providing the 5V yields the following:
Note that the hash around 1 KHz is gone. We can see from the RMS figures that the noise floor has improved about 0.5 dB using the lab supply.
And how about a no-name USB hub?
Finally, below is a plot running from an "Anker" USB3 hub powered by a "Flypower" brand DCDC external switching power supply (aka wall wart). It is virtually indistinguishable from the lab supply.
Conclusion
The convenience of USB power is hard to beat, but it can come with a cost in some cases. If you are making crucial measurements near the noise floor of the device, then spend some time to acquaint yourself with the USB supply sensitivity of your setup. In most cases, it won't matter at all even if you are using a no-name power brick to power your hub. But in some cases, you could see some noise introduced into your measurements if your USB power is exceptionally poor.
The solution is pretty simple if you cannot find a clean USB port on your machine: Just use a USB powered hub when you need to squeeze the last few tenths of a dB from your measurements.