Developing a Regulated Dual Rail Power Supply For FirstWatt Amps

Modified circuit to give flat 25mOhm output impedance

Sizi%20Supply.png





Output impedance

Sizi%20Supply%20output%20impedance.png
 
For reference, for the stock ccrcc supply (15000u, 15000u, 0.12 Ohms, 15000u, 15000u) here is the impedance.
I don't know if I am just super anal about this (maybe just a bloody storm in a teacup) but I don't like the rising impedance response below 100Hz where there is a lot of bass energy.


CCRCC%20Impedance_1.png
 
Member
Joined 2011
Paid Member
If the amplifier dissipates 180 watts during operation, as the Firstwatt F6's owners manual states, and if the raw unregulated supplys are ± 30 volts as shown in the simulation files here, then the DC bias current is about 3 amps. The amp is rated for 50W into 4 ohms so that's an RMS output current of 3.5 amps and a peak output current of ± 5 amps. The PSU idles at 3A and is occasionally called upon to deliver 5A.

If the amplifier drives a full size square wave into a 4 ohm load, the output current of the PSU switches between 1 amp (it's a class A design so the current never falls to zero) and 5 amps. Delta-Current is 4 amps. If the PSU output impedance is 25 milliohms then the ripple on the "regulated" supply is 25*4 = 100 millivolts. Multiply this by the amplifier's PSRR, which is unlikely to be enormous for these kinds of common source amplifiers with very modest amounts of global feedback, and you get the equivalent input signal that a 100mV ripple on the "regulated" supply produces. Have your engineer run the numbers and see if she's pleased.
 
If the PSU output impedance is 25 milliohms then the ripple on the "regulated" supply is 25*4 = 100 millivolts. Multiply this by the amplifier's PSRR, which is unlikely to be enormous for these kinds of common source amplifiers with very modest amounts of global feedback, and you get the equivalent input signal that a 100mV ripple on the "regulated" supply produces. Have your engineer run the numbers and see if she's pleased.

Yes I understand all that but it's a lot better than 100mOhms of the stock supply.

What suggestions would you make following the KISS principle?
 
Member
Joined 2011
Paid Member
Gain and feedback will lower the impedance and increase the line rejection ("ripple reduction") at DC and at low & midrange frequencies. Some people are frightened of gain and feedback.

All you need to do is choose a target and then hit it.

For example you could build an error amplifier out of discrete components, whose gain-bandwidth-product was 10 Megahertz and whose gain was 20X. This will reduce your output impedance by (1 + gain), namely, 21X. From 17 milliohms to 0.81 milliohms.

Since your error amplifier's GBWP is 10 MHz and its since its gain is 20X, its open loop gain rolloff frequency is 500 kilohertz. So your output impedance will be ruler flat from 0.001 Hz to 500 kHz. Yes you can tell SPICE to begin its AC analysis sweep at 0.001 Hz. Whether "flat Zout=0.8 milliohms from DC to 500kHz" hits the target you have set for yourself, only you can say.
 
Last edited:
Hi Mark
Once we add say 5 mOhms in wiring and contact resistances etc
That rise in response is going to look a lot flatter than it looks now.
Do you think it's important to keep squeezing more performance out of the circuit given the impedance will be subject to cabling and connections all with their own resistance values?
 
Member
Joined 2011
Paid Member
Yes it appears that AWG16 stranded copper wire has a resistance of 4 milliohms per foot so if your connection between PSU and AnalogPowerAmp is six inches long, that's an unavoidable 2 milliohms of output impedance. Then I suggest that you make the PSU's output impedance one fourth of that (500 micro ohms) and focus your energy upon maximizing the bandwidth.

Post #35's plot shows an output impedance of 23 micro ohms and a corner frequency of 16 kilohertz. If you decreased the error amplifier's gain by 20X while keeping its gain-bandwidth product the same as it is now, you'd get an output impedance of 460 micro ohms {exactly as targeted in previous paragraph} and a corner frequency of 320 kilohertz. Which would be excellent in my opinion.