Until recently, I had been quantifying Tiliqua’s analog front-end with an oscilloscope and my ear. Unfortunately, using my ear is slow and oscilloscopes don’t have the dynamic range to measure key audio specs like noise floor or THD accurately, as their ADCs are often limited to 10 or 12 bits (and we want to measure a 24-bit audio CODEC).

This week, as part of prepping for more automated testing (for production), I got my hands on a high quality audio analyzer. In this update, I’ll share my experience measuring Tiliqua’s audio front-end with it and compare that to another low-noise Eurorack module.

The Setup

For the lowest possible noise floor, it is common in audio to use balanced (differential) signaling, where one wire carries an inverted version of the other wire. As Eurorack is inherently single-ended, I had to terminate the negative inputs of the audio analyzer, as you can see here:

This kind of setup is not ideal, but as long as using the analyzer in single-ended mode does not introduce more noise than the device we are measuring, it should not be a problem. To make sure everything was working, I first ran a tone through the two jacks connected together in loopback, to see what we’re up against:

In the hypothetical case that our audio front-end is absolutely perfect, it should be possible to reach the specs of the CODEC chip. From the AK4619VN datasheet, assuming we’re running 48 kHz single-ended, the ADC is specified for a THD+N of 95 dB typical and the DAC is specified at 91 dB typical. As both of these values are much higher than the 110 dB shown above in our setup, we can make some measurements!

Measuring a Disting

Before measuring Tiliqua, I measured quite a few digital modules in my rack to try and find a good comparison. In the end, the lowest noise digital module that I tested was a Disting mk4. Ironically (and perhaps unsurprisingly), this is one of the few modules where specs like SNR and THD+N are actually provided by the manufacturer. Anyway, here it is hooked up:

After putting the Disting in passthrough mode (well, precision adder mode) and switching on the test tone (whilst being careful to avoid any clipping), we have a THD+N around 86 dB:

On the manufacturer’s website, the official spec is -93 dB for the ADC and -94 dB for the DAC. Since we’re running in loopback, with two inputs summed together by the Disting, we are really measuring the THD+N of 2 ADCs and the DAC summed together (extremely roughly: -93 dB + -93 dB + -94 dB = -88.5 dB). Not too far off! The 2 dB difference might be from using a stimulus that is not full scale for the Disting’s ADCs/DACs.

Measuring Tiliqua

Next it was Tiliqua’s turn. Here I hooked it up in exactly the same way:

Before testing, I flashed the mirror bitstream, which sends all inputs to outputs, using an internal sample rate of 48 kHz (Tiliqua can of course handle 192 kHz, but 48 kHz has slightly better noise performance):

This shows some 50Hz harmonics at the low end, perhaps coupled in from my lab PSU.

Interestingly the THD+N of 87 dB is slightly better than the Disting, however, we have to take into account that we are only measuring one ADC+DAC pair in loopback, not two ADCs as before. So on this metric at least, a single Tiliqua channel is ever so slightly worse than a single Disting channel, but there is really not much to it (1-2 db), especially considering how much high-frequency, high-current activity is happening on the Tiliqua motherboard nearby!

Comparing with the best-case-scenario: adding up the typical CODEC performance from the datasheet (-95db ADC + -91 dB DAC = -89.5 dB typical), our measurement is pretty close. Not bad!

Amongst other things, I also took a closer look at Tiliqua’s gain flatness and crosstalk between channels:

Gain flatness on stimulation by white noise

Crosstalk between adjacent audio channels (-93 dBV-11dBV = -104 dB of rejection)

Fortunately, in all cases there were no surprises. Subjectively, at least, even with the volume cranked up to 11, noise from Disting (and Tiliqua) is essentially inaudible through headphones, and more than acceptable for Eurorack. In Tiliqua’s case, I was never able to hear any undesired coloring of the sound, but I am glad that the measurements support my ears!

Future: Automated Testing

Eventually, my goal will be to use this audio analyzer as part of Tiliqua’s hardware testing/QA procedure to make sure every unit performs identically, without needing to do comprehensive listening tests on every unit (with my ear). I’ll be sure to post some pictures of this when the time comes!

Bonus: Color Modulation

In the background, I’m continuing to add more features to the example bitstreams as fast as I can. This week I was having some fun with xbeam‘s color modulation feature, I’ll just drop a picture, more coming soon!

Thanks for Your Support!

As a reminder, our campaign is still live with less than 2 weeks to go! If you haven’t yet, reserve your Tiliqua now while it’s available at our lowest campaign price!
If you have questions of any kind about Tiliqua, feel free to contact us by email, using the "Ask a question" link on our campaign page, or on Matrix. Until next week!