In designing DACs I've figured out from listening that I need to strive for the lowest possible noise in my circuits, even well below the noise on a 16bit recording. This was rather a big surprise at first, but how low to go in noise before lowering it further makes no difference?
In order to get a handle on this question its interesting to consider how good an SNR can be got from simple IC (opamp) circuitry - the lowest noise opamp I've been using is LT1028 which has input noise around 0.9nV/rtHz. This corresponds to noise at the output around 130nVRMS. Hence relative to a nominal DAC output level of 2VRMS an SNR of 144dB. In practice though there are normally resistors in the circuit and they'll contribute noise beyond that of the opamp, given the LT1028's noise is roughly equivalent to a 50ohm resistor. A 50ohm working impedance in the circuit isn't really practical as it'll call for significant drive current, beyond the capabilities of an unassisted opamp.
Rather than chase noise lower there's another approach - push signal levels higher. But then we run up against the 2V output standard - if our DAC outputs a maximum level higher than this it won't be widely compatible with existing audio kit. But if there were a low enough noise means of attenuating the output of a higher voltage DAC then there could be a route to running much higher signal levels within the DAC. It seems to me there is only one solution - an autotransformer as a simple resistive attenuator will either be power-hungry to drive if low noise or too noisy if practical to drive. Since an autoformer is in effect a tapped inductor the impedance seen by the active circuitry is that of an inductor whereas the noise level will be that of the copper winding resistance.
Having wound transformers in the past using ferrite cores (because they're cheap and easily available) I decided to look into what impedance (and hence noise) might be achievable. I have a box full of RM12 cores left over from an earlier DAC project, PC40 material which is good to about 0.35T, resulting in 185 turns per volt to handle 20Hz. I found I could fit about 2100 turns of some 0.16 outside diameter wire on. So about 11.3V peak at 20Hz. The resistance end-to-end was ~100ohm and the inductance just a smidgeon below 22H, corresponding to j2700ohm at 20Hz. So not a big challenge to drive, just over 4mA peak current.
As a 12dB attenuator the resistive portion of the output impedance was in the region of 25ohm thus an SNR of 147dB should be achievable if the preceding electronics was completely noise-free. A promising start I reckon. With signal levels 12dB higher even a lowly OPA2209 can achieve an SNR around 148dB (excluding resistor noise).
Having an autoformer on the output of the DAC makes me think I should use that as the primary volume control in the system. So that's the next design challenge.
In order to get a handle on this question its interesting to consider how good an SNR can be got from simple IC (opamp) circuitry - the lowest noise opamp I've been using is LT1028 which has input noise around 0.9nV/rtHz. This corresponds to noise at the output around 130nVRMS. Hence relative to a nominal DAC output level of 2VRMS an SNR of 144dB. In practice though there are normally resistors in the circuit and they'll contribute noise beyond that of the opamp, given the LT1028's noise is roughly equivalent to a 50ohm resistor. A 50ohm working impedance in the circuit isn't really practical as it'll call for significant drive current, beyond the capabilities of an unassisted opamp.
Rather than chase noise lower there's another approach - push signal levels higher. But then we run up against the 2V output standard - if our DAC outputs a maximum level higher than this it won't be widely compatible with existing audio kit. But if there were a low enough noise means of attenuating the output of a higher voltage DAC then there could be a route to running much higher signal levels within the DAC. It seems to me there is only one solution - an autotransformer as a simple resistive attenuator will either be power-hungry to drive if low noise or too noisy if practical to drive. Since an autoformer is in effect a tapped inductor the impedance seen by the active circuitry is that of an inductor whereas the noise level will be that of the copper winding resistance.
Having wound transformers in the past using ferrite cores (because they're cheap and easily available) I decided to look into what impedance (and hence noise) might be achievable. I have a box full of RM12 cores left over from an earlier DAC project, PC40 material which is good to about 0.35T, resulting in 185 turns per volt to handle 20Hz. I found I could fit about 2100 turns of some 0.16 outside diameter wire on. So about 11.3V peak at 20Hz. The resistance end-to-end was ~100ohm and the inductance just a smidgeon below 22H, corresponding to j2700ohm at 20Hz. So not a big challenge to drive, just over 4mA peak current.
As a 12dB attenuator the resistive portion of the output impedance was in the region of 25ohm thus an SNR of 147dB should be achievable if the preceding electronics was completely noise-free. A promising start I reckon. With signal levels 12dB higher even a lowly OPA2209 can achieve an SNR around 148dB (excluding resistor noise).
Having an autoformer on the output of the DAC makes me think I should use that as the primary volume control in the system. So that's the next design challenge.