A month ago, in August, I posted an update to my original USBDAC post. I had ordered the prototype revision B PCBs and they arrived a week ago or so. I assembled a test unit but it didn’t work. I debugged the circuit and asked around for help and quickly found a problem; VDD (3.3V) on the circuit does not equal VBUS (5V). Cutting the VBUS to VDD trace didn’t help though, still nothing when I plugged in the USB cable. After some more debugging and circuit referencing I found two more problems. My USB D+ pull-up was 1 megaohms instead of the 1.5 kilo-ohms specified, and it was pulled up to VBUS instead of VDD. After fixing that, Windows actually recognized the Composite Device (the PCM2706 presents itself as an Audio class device as well as a HID class device) but it immediately said both the HID and Audio devices were “unplugged”. The Windows “device unplugged” notification would sound every 10 seconds or so.
After deciding to ditch board #1, I took another one from the batch. All the modifications mentioned above were made and I soldered all the digital side components on and this time it worked! I had no analog components on, but I verified with the scope that sane audio signal was actually being output from the chip. It worked after soldering on the analog side components as well and, to say the least, I was quite thrilled.
(The whity residue on the board is flux. Use generous amounts like I did. It cleans off nicely afterwards.)
Initially I tried it with cheap in-ear headphones and it’s loud. The Windows mixer was set to 5/100 and it was still quite loud. I listened to Spotify a few moments with external Creative speakers plugged in and everything worked nicely.
Today I brought #2 to work with me to have a listen with a bit better pair of headphones (Beyerdynamic DT 331). After a few hours listening I can tell that the highs and mids are really crisp and “articulated” but bass is lacking. Seriously. I don’t know if it’s because of the analog output circuitry or what, but I’m quite disappointed at that. My reference setup is the same pair of headphones, driven by Edirol MA-10A monitor speakers’ headphone out, fed by my work computer’s built-in “Realtek HD” sound card.
There’s just one problem that I have to fix for “production” release; the channels are swapped. But that’s an easy fix. (And the VDD/VBUS and D+ pull-up issues as well, naturally.)
I have to google around and see what kind of output stages people usually use with these and experiment. I suspect that better quality electrolytic caps could help as well. The ones I’m using are just ordinary filter caps by Panasonic. A more stable and accurate oscillator would be in order as well. Even with my low-end Rigol oscilloscope I was able to measure that the oscillator is running at 12.005 MHz, so that’s already 5 kHz too fast. A basic TCXO should do, and it only costs a few euros.
Update: Found the problem to sucky bass response; I had somehow managed to read the datasheet’s reference design wrongly — again. *Sigh* Turns out the final RC high-pass filter on the output stage is supposed to have a 100 μF cap and a 3.3 kOhm resistor (Fc of ~0.5 Hz) but I somehow specified 100 μF and a 16 ohm resistor. That means that the cutoff frequency is 100 Hz, nicely cutting out all the bass. Fourtunately it’s a nice quick fix to switch them over to 3.3 kOhms. I hope it remedies the situation.
(How fucking hard can it be to read plain, legible English? I made about a dozen mistakes interpreting the datasheet. Well, this is how you learn electronics. Make mistakes. A lot of them.)
Update #2: I switched left channel resistor to 3.3 kOhms like it should be and oh boy does it sound good now! Connected to NAD C320BEE and Amphion Helium²s it really comes alive. The bass kicks nicely but all the mid and high clarity is retained. Even when I crank the NAD to full volume with nothing playing, I can’t hear any noise. The golden silence.
I’m updating the schematics now with the modifications mentioned above, plus I’ll be changing the analog side Panasonic aluminum electrolytics to premium audio-grade Nichicon MUSE F95 solid tantalum electrolytics. I will also change the crystal to temperature-controlled crystal oscillator (TCXO). I think that should complete the package. :) I expect the Rev C to be ready sometime in September or even October.
(If you’re interested in buying one of the revision C units, contact me!)