Designing a linear regulator with good "line regulation"

[dzseki]

I was given a task to design a power supply for an equipment. The most interesting outputs are analog +/-5V (max. 1A) and analog +/-15V (max. 500mA). I wanted to keep it simple and used off the shelf 3pin regulators (78/7915 for 15V and LM2940/2990 for the 5V) and built a prototype board (attached the schematic excerpt). The guys who would use this power supply are not very happy with the performance of the circuit so far. They would expect to have ripple on the output below 1mVpp at rated load. Instead, for example on the +15V line I have 4.4mVpp ripple right now (see picture). On the +/-5V lines the perofrmance is somewhat better with ripple around 2mVpp, but the voltage is also lower so that is not a big surprise.
So how could I improve this?
-I could put bigger filter caps by one size 2200uf->3300uF but I think that would only slightly improve the performance.
-What I can’t decide is whether to add preregulator circuits before these regulators or should I attempt to design an own regulator having better line regulation in one step?

For preregulation I’d use again 3pin regulators or zener driven emitter follower. For discrete regulator I was thinking about a well regulated voltage reference (one for the + side and one for the – side) and using error amplifiers with feedback off emitter follower output.

The size of the circuit board is 130x200mm and has 10 power devices already needed to be heatsinked, adding preregulators would increase this to 14, and the circuit layout is very dense heatsinking would be very difficult. But having four diescretely built regulator would also take up PCB real estate...
The power supply does not have to do everything right, eg. the voltage temperature stability may be traded for better line regulation, also over current limitation is not so critical (eg. polifuse is sufficient).

Hint for the scope shot: yellow trace is the input voltage to the regulator (ie. filter cap voltage) blue trace is the output ripple of the regulator (AC coupled) all at 500mA load.

[madires]

Have you checked out the regulators' datasheets? Both voltage regulators can't do much better. You could tweak the output caps to improve things slightly, but I'd guess it won't be enough. BTW, why is C17 10µF and C18 100nF? IME the 78/79xx like a 100nF plus a small electrolytic. What about a protection diode for each regulator from output to input? A preregulator wouldn't change much as it doesn't improve the regulator's line and load regulation. Maybe someone else can suggest voltage regulators with better specs.

[Conrad Hoffman]

It's very easy to mess up the pcb layout and get ripple on the output. Look carefully at the cap charging current loop and be sure it has no paths in common with the output loop. I've also had excellent results from the LM317 regs compared to others. You can filter the ref pin with those.

[DaJMasta]

I'd agree the output capacitance looks a bit low, maybe the 47uF caps on the 5V rails is reasonable, but 10uF on the 7915 seems on the low side and 100n on the 7815 is definitely small for what I've seen/used.  Fusing the output of the linear reg and not the input (or input of the rectifier) is later than I would use and its DCR could be making some ripple on the rail if there's regular current consumption spikes, though it would do nothing for a constant load.

It may be worth using the FFT function on your scope to look at the frequency content and see if it corresponds to frequency content on the input lines.  If it does, some common mode filtering on the AC side of the rectifier (or maybe before the transformer) could help out.

[dzseki]

More careful grounding is a good idea, I'll try.
The 100nF capacitor was just a starting value, when I made the measurements a 100uF tantallum capacitor was already added to the output, seemengly it did not change much though.

[floobydust]

Make sure you use single-point grounding, at the filter caps as they have high ripple current. Where is your power transformer center tap wire to?
Your 500mA fuse will have high DC resistance and is better off located at the input side of the Vreg.
I went from 15mV to 1mV ripple just moving some connections with 3-pin voltage regulators.

[dzseki]

I've also had excellent results from the LM317 regs compared to others. You can filter the ref pin with those.

Indeed, I've put together a quick rats nest style LM317 regulator with the same AC-to-DC frontend and got ripple well below 1mV, with 10uF filtering on the ADJ pin.

[namster]

i think that for reduce the ripple on output first you have to take in consideration the PSRR of regulator for example the 7815 have minimum value of 60dB
then we have to calculate  the   expected Ripple Input
PSRR = 20Log (Ripple Input)/(Ripple Output)
PSRR/20=Ln((Ripple Input)/Ripple Output))/ln10
exp(ln10*PSRR/20)=(Ripple Input)/Ripple Output)
Ripple Input=Ripple Output*exp(ln10*PSRR/20)
Ripple Input = 1mVpp*exp(ln10 * 60/20) = 1V
after that you can chose the right filtering capacitor by using this equation
Vpp=I/2*f*C C=Vpp=I/2*f*Vpp      C= 5000µF you can take a normalised value of 6800µF

[dzseki]

While true, I find this a  less elegant approach.

[namster]

why is it an less elegant approach ?

[dzseki]

Because a 6800uF capacitor is about 3-4 times bigger (and more expensive) than a 2200uF one, cascading two 78xx regulators would give 120dB PSRR for less than 2USD, but I had little PCB area left for even that. And as said above the LM317 does the trick as well basically at no extra cost and circuit space.

[namster]

cascading two 78XX isn't a good idea , with a Minimum Droupout Voltage of 3V the two regulator will dissipate 3Watt , and i'm happy that you solved the problem .

[dzseki]

cascading two 78XX isn't a good idea , with a Minimum Droupout Voltage of 3V the two regulator will dissipate 3Watt , and i'm happy that you solved the problem .

Right now the unregulatted voltage before the regulator is 22V so that "excess voltage times current" has to be dissipated anyway.