Filter for switch mode power supplies

[tomp]

Music server power supply filter.pdf

(19.09 KiB) Downloaded 703 times

I previously mentioned I would be making a filter for the output of the switch mode power supply that feeds the Intel NUC that acts as my music server. There has been a lot of discussion about building or purchasing sometimes large and expensive linear power supplies for music servers. I wanted to see what would happen if I retained the standard power supply and added a passive filter between it and the NUC. I previously mentioned it but I have now progressed further so I decided to open a new thread.

The concept is to build a filter that would effectively decouple the NUC from any noise generated by the power supply. To achieve that, there is an initial stage of high frequency filtration, followed by a large choke, and then several stages of output filtration including a very large storage pool of capacitance in stages. To insure that the power supply would start into that very large capacitive load, a P channel enhancement Mosfet was used in series with the large filter bank on the output to provide a soft start function.

Another consideration since the filter was passive was to minimize any voltage loss across the filter. TO do that, a choke with very low DC resistance was chosen with a value of .05 ohms. It is rated at 2.5 mh at 10 A. The Mosfet used also had an on resistance of about .02 ohms. The following is a circuit description:

Input filter capacitors C1 - C10 are 0.1 uf monolithic ceramic which are very good at bypassing high frequency signals. L1 is the choke mentioned. C24, C25, and R1 form a timing circuit to slowly turn on the series Mosfet. Q1 is the Mosfet that starts out at a very high resistance but as the capacitor C25 charges, lowering the gate voltage on Q1, it turns on slowly, eventually getting tho a very low resistance saturated condition. C11 is a large 10,000 uf electrolytic capacitor that provides the bulk of the filtration on the output and in conjunction with the choke forms a very effective low pass filter. C12 is a 1,000 uf electrolytic that has a series 0.5 ohm resistor to provide a step function to the filtration bank and also some damping. C13 is a 1 uf film cap that further provides a continuation of the filtration function of the output circuit. Output capacitors C14 - C23 are 0.1 uf monolithic ceramics that not only further reduce any high frequency noiose that may get through the filter but also act as a noise sink to any noise present on the power input from the NUC itself.

As far as voltage drop across the filter, when connected between the power supply and the NUC with the NUC running and playing music, the input voltage to the filter from the power supply measured 19.65 volts, and the output of the filter going to the NUC was 19.60. Certainly with the current beng drawn by the NUC under actual operating conditions, the voltage drop is only 50 mv which is totally inconsequential.

I took measurements of the noise going to the NUC without the filter and then with the filter, both under actual operation conditions. In both cases the time base of the oscilloscope was set to 500 us per division and the vertical sensitivity was set to 50 mv per division. You can see that there are many high frequency spikes in the power going to the NUC with peaks reaching just under 200 mv. After running through the filter, with the same settings on the scope, the trace is basically a straight line with what looks like a very low level of white noise. These traces were made on my work bench with an open circuit with no benefit of shielding from a metal chassis.

At this point, I have not done any listening tests with the filter in circuit as that will have to wait until I get it in an enclosure. My next post will have some construction photos and the results of the listening test. I'll also include parts costs. Looking at the measured results I would say the project so far is a success in eliminating the noise coming from the switcher, making it look to the NUC as a linear supply. You might want to consider this as an option to purchasing or building an expensive linear supply.

Tom

Attachments

Server power direct.pdf

(272.46 KiB) Downloaded 749 times

Server power filtered.pdf

(163.97 KiB) Downloaded 685 times

[Pelliott321]

Sounds interesting
Pleas include schematic

[tomp]

The schematic was included at the top of the post.

[Pelliott321]

Sorry I missed that.
The multiple .1 caps are just to lower ear?

[TubeDriver]

Interesting approach Tom. I went with a linear PS in part because I was more more concerned about noise migrating back into the other components in my system. The outputs of my balanced power surge/line conditioner are not isolated from each other and I have both my streamer and DAC plugged into it. So it seemed like the main risk was switching PS noise traveling into my DACs PS. I think the answer is yes, but will your filter work in that direction too?


It sounds like your filter is a couple diodes away from being a linear PS, any reason you wanted a filter/switching PS instead of just going with a linear PS?

[Roscoe Primrose]

It sounds like your filter is a couple diodes away from being a linear PS, any reason you wanted a filter/switching PS instead of just going with a linear PS?

I can see two good reasons for that...

1. He already had the switching supply.
2. Switching supplies are much more efficient.

Roscoe

[TubeDriver]

All true but the NUC is using about 7.5w, not exactly a power hog. My concern about switching PS is less for the Nuc and more for everything else plugged into the same line.

It sounds like your filter is a couple diodes away from being a linear PS, any reason you wanted a filter/switching PS instead of just going with a linear PS?

I can see two good reasons for that...

1. He already had the switching supply.
2. Switching supplies are much more efficient.

Roscoe

[tomp]

Sorry I missed that.
The multiple .1 caps are just to lower ear?

The multiple caps are there because I have lots of them and it is a good way to get higher capacitance without increasing series impedance when using a single cap. These components are like so many that the freight cost is more than the component so usually I buy things like these caps, small diodes, resistors, etc. in bulk.

[tomp]

Interesting approach Tom. I went with a linear PS in part because I was more more concerned about noise migrating back into the other components in my system. The outputs of my balanced power surge/line conditioner are not isolated from each other and I have both my streamer and DAC plugged into it. So it seemed like the main risk was switching PS noise traveling into my DACs PS. I think the answer is yes, but will your filter work in that direction too?


It sounds like your filter is a couple diodes away from being a linear PS, any reason you wanted a filter/switching PS instead of just going with a linear PS?

My general philosophy is that each device in the signal chain had better have a well enough designed and built power supply to stand on its own and not be affected by line noise regardless of where it comes from. Even a linear supply should have the kind of filtration I show in the filter. All the signal level stuff I build has a similar choke filtered supply after the input voltage regulators. Having a weak sister somewhere in the chain is the point to address, not try to make all the other devices try to attain a very high standard just because of that one weak component. I'm including a schematic of the power supply I use in my preamp which is Op Amp based. It is a linear supply but just because it has a linear front end does not make it immune to line noise getting through the transformer because of interwinding capacitance, So after the regulators I have the same type of LC network that forms a passive low pass filter that is very effective and provides a huge current reservoir for the amps. BTW, the type of filter I showed above will work for any DC powered device although it may have to be bipolar as shown in the preamp schematic.

Tom

Attachments

Preamp power supply.pdf

(24.14 KiB) Downloaded 664 times

[tomp]

I had just a little time to listen since I put the filter in line with the NUC power supply but here are the results. I was expecting little to no effect when I used the filter. Actually I was very surprised at the difference it made. As I listened I kept saying this can't be happening, but it was. Here are some specifics on a few test tracks I used:

Magnificat recorded by MSM Studio. It is a recording the music of Arnesen of a large choral group in a church in Norway accompanied by an organ. There are 60 people in the chorus distributed between sopranos and tenors. Not only is the hall sound spacious, but the intensity and mix of the voices is very hard to separate without sounding hard. The filter displayed a texture in the voices and a smoothness I had not heard before. Listen to track 5 Fecit potentiam.

Garth Brooks - Friends in Low Places. This live recording can result in the audience voices becoming congested and hard when played back at high levels. There was no sense of that after using the filter.

Verdi - Aida - von Karajan - si core voce. This cut ranges from single voices at low levels to high levels of massed voices. It was recorded in a spacious venue and things have to be correct to get the sense of space. If things are not right the voices will get congested very quickly. The space and voice separation improved with the filter.

Billy Idol - White Wedding. Again a live recording where the audience is very enthusiastic. The filter greatly helped in the separation of the voices even at high levels with no sense of hardness and the applause was uncanny.

Clark Terry - Live at the Village Gate - track 8 Hey Mr. Mumbles. This is a superb Chesky recording with a very active live audience. The ability to place the individual audience voices in space using the filter was uncanny. Also things like the mute on Terry's horn has just the right timber and bite without being harsh.

Gershwin - Rhapsody in Blue. This is a recording a friend who has recorded for NPR did in a venue in Colorado and sent to me. He uses two DPA measurement mikes in a spaced array with a center fill. For this recording he had another mike on the piano for additional highlighting of that instrument. The filter made the piano sound very realistic without excessive hammer sounds on the strings as so happens when a mic is used on the piano. It sounded like a real piano.

So my original assumption of the filter not making a significant difference was proven totally wrong. That is why I test. I always use measurements to start with as I did in this case, but the measurements do not always correspond with what I hear. To say I was very surprised is an understatement. I can't begin to explain why. I'll leave that to others who are deeper into digital sound reproduction.