Years ago I built a USB DAC audio interface for my brother. I had intended it for one of my aunts but it refused to work with her laptop, so my brother got it.
This was based around the TI PCM2706, SRC4192 and PCM1794. Time has moved on and while the PCM2706 gave a lot of, easy to implement, functionality it's been significantly left in the dust by more modern designs. Having used the CM6632A in another project I decided to build a replacement unit for him.
To make the replacement worthwhile I wanted to improve upon every aspect of the previous design, both in terms of performance and ease of use. Most notably the previous unit required a separate AC power feed which made it cumbersome to move around. In this replacement unit the entire thing runs directly from the USB cable.
Power mangement and the DAC
As typical USB ports are only specified to provide 500mA, at 5V, this places constraints on the design.
The first issue is the requirement for a split supply for the opamps. Not only would a negative rail be needed but the 5V from the USB rail isn't enough voltage either. This requires a two pronged attack of inverter and boost converter. Luckly the engineers at TI are smart people and have foreseen this kind of situation. They make a handful of chips designed to do just this. Here I chose to use a TPS65131.
One of the traits of the PCM1792A, current output DAC, is that it biases its outputs so that they are always sinking current. This happens regardless of the signal level and when the DAC is outputting all zeros the outputs sink 6.2mA of current. This then swings down to -10.1mA and up to -2.3mA when the output is at full scale. As a result of this, after I/V conversion, the entire voltage swing of the output is offset below ground, within the negative portion of the opamps output range. Of course after the I/V stage comes a differential summing stage. This cancels out the differential negative bias on the I/V outputs and centres the output about ground.
As the I/V stage operates entirely within the negative portion of the opamps output the supply to said opamps needs virtually no positive rail swing at all. Only needing just enough positive potential to ensure that the output swing limits of the opamp are respected so that the signal doesn't clip. The negative rail, on the other hand, needs considerable head room. As the I/V stage requires little in the way of positive rail voltage the requirements for the positive rail fall on the opamps that perform the differential summation. Their output swing is centred around ground so they require a significant amount of voltage swing into the positive direction.
Primarily the output of the DAC is to be used to drive a pair of headphones and, in this case, I chose to use an INA1620, from TI. This opamp sums the differential signal on the output of the I/V stage and drives the headphones at the same time. Whatever lots of people may try and tell you, large voltage swings are not usually necessary for driving high impedance headphones. My brother uses a pair of HD650s and these can reach blisteringly loud volumes without needing insane voltage swings. I too have a pair of 650s and can tell you they go plenty loud enough with only 2VRMS of voltage. To that end I designed the analogue circuitry, after the DAC, to achieve a voltage swing just in excess of 2VRMS.
Looking at the INA1620s datasheet it says that with a 600ohm load it can swing to within 1V of the rail voltage. With lower impedances this would be even lower, but given a 2VRMS signal only needs to swing to a peak of ~2.8V, powering the opamp from a 5V positive rail should provide more than enough headroom for a full scale output from the DAC.
Ironically what ended up setting the maximum requirement for the positive supply rail wasn't the opamps at all, rather it was the drop-out of the 5V LDO, necessary, for the DACs analogue circuitry. While LDOs may be able to operate with only a small drop between their input and output their performance usually improves when provided with at least 1-2V of headroom. Here the LDO used was the ADP7118, from Analog Devices, and I chose to set the positve output from the boost converter to 6.5V.
With the component values selected for the I/V stage and the NE5534 opamps used, setting the output of the inverter to -12V worked very well. Post switching converter regulation being provided by a TPS7A39 dual, split supply, LDO.
The main advantage of using asymmetric power supplies, for the analogue circuitry, being that it dramatically reduces the power consumption. Usually this isn't important for most hifi applications, but here, it was necessary.