A 'brick wall' filter for the +DC rail of a SMPS?

[andyr]

My issue:
* I have an AC motor speed controller on my TT which is powered by a +48v Meanwell SMPS.
* I did an experiment where I used a 48v SLA supply to power the speed controller, instead of the Meanwell.
* The music sounded better as a result!

People have suggested this is because a SMPS has HF hash on the +DC output ... and this has a negative effect on SQ.

Obviously, I could use SLAs on a permanent basis for the AC motor speed controller but this would be quite an expensive solution (~$500, including batteries, case, charger etc.).

Someone suggested a brick wall filter (at, say, 2 or 3KHz) on the DC output of the SMPS would solve the problem - as this would stop the HF hash from getting into the speed controller. This would certainly be much cheaper than $500 - so I was hoping someone could tell me how to implement such a brick wall filter.


Thanks,

Andrew

 

[abraxalito]

Brick wall filters only exist in theory, real filters are somewhat less of a 'wall'. You can start out by going here - Chebyshev Pi LC Low Pass Filter Calculator

Notice that you have to have a working impedance - something like 0.1ohm would be a practical first stab.

There is also the possibility that common-mode noise from the SMPSU could be coupling into your audio. A filter on the DC output won't deal with this.

 

[andyr]

Brick wall filters only exist in theory, real filters are somewhat less of a 'wall'. You can start out by going here - Chebyshev Pi LC Low Pass Filter Calculator

Notice that you have to have a working impedance - something like 0.1ohm would be a practical first stab.

Yes, I understand that "real-life" filters are not as steep!

Thank you so much for posting that link. I have played around with that calculator and have a couple of Qs, if you don't mind:

1. Does the slope get steeper the more "pis" you use? IE. a pi filter using 7 components will be steeper than one which only uses 3 components?

2. For a 7-component pi filter, the calculator gives the 2 cap values as 940uF & 1,668uF. Do these have to be exact ... or could I instead use:
* 1,000uF, and
* 1,000 + 3x220 = 1,660uF?

With a 63v rating, given the +48v DC supply.

There is also the possibility that common-mode noise from the SMPSU could be coupling into your audio. A filter on the DC output won't deal with this.

Quite right but I have already dealt with this by putting a mains isolating transformer between the SMPSU and the wall socket. (This sounded much better compared to not having the isotran ... but the battery PS sounded best of all! )

Thanks again,

Andy

 

[abraxalito]

Thank you so much for posting that link. I have played around with that calculator and have a couple of Qs, if you don't mind:

1. Does the slope get steeper the more "pis" you use? IE. a pi filter using 7 components will be steeper than one which only uses 3 components?

Questions are one of the biggest reasons for hanging out on DIYA for me Yes, the more elements (components) in the filter, the steeper the slope gets. Eventually though real-world considerations like capacitor ESR and inductor ESR intrude (they're not accounted for in this calc).

2. For a 7-component pi filter, the calculator gives the 2 cap values as 940uF & 1,668uF. Do these have to be exact ... or could I instead use:
* 1,000uF, and
* 1,000 + 3x220 = 1,660uF?

No, they certainly do not need to be exact, the main variation with values won't much concern you as it affects passband flatness. In a DC application the passband can be arbitrarily narrow. Given that electrolytics are at best only 20%, I'd not bother to go even as close as 1600uF - try at first with all the same values and see where it gets you.

There is a hidden gotcha here - the filter only works as advertised if terminated with 0.1ohms. Given that'll effectively short your SMPSU out if just a 0.1ohm resistor, it needs to be an AC termination. For that you'll need a capacitor which typically is more than 10X the size of the output capacitor calculated by the webpage. The 0.1R resistor goes in series with this big cap.

For this reason its probably worth experimenting with slightly larger than 0.1R termination - it'll give more manageable termination cap values.

Quite right but I have already dealt with this by putting a mains isolating transformer between the SMPSU and the wall socket. (This sounded much better compared to not having the isotran ... but the battery PS sounded best of all! )

Good move but all transformers have interwinding capacitance - do you know what the value is for yours? You may find that CM noise still manages to get through.... Custom designed CM chokes can probably improve on the AC isolation given by your trafo.

 

[andyr]

There is a hidden gotcha here - the filter only works as advertised if terminated with 0.1ohms. Given that'll effectively short your SMPSU out if just a 0.1ohm resistor, it needs to be an AC termination. For that you'll need a capacitor which typically is more than 10X the size of the output capacitor calculated by the webpage. The 0.1R resistor goes in series with this big cap.

Sorry, abraxalito, you have confused me - perhaps you will bear with me.

Instead of the +48v DC output going to the PCB of the motor speed controller PCB ... we are now going to have this pi filter in between the SMPSU and the speed controller PCB.

So, doesn't the speed controller PCB "terminate" the pi filter? (So I just have to find out what Zin of the speed controller PCB is.)


Regards,

Andy

 

[abraxalito]

No problem

The filter designer designs filters which have equal input and output impedance. The speed controller PCB would need to have a 0.1ohm input impedance which seems very unlikely indeed. It'll be much higher than 0.1ohm in practice hence we can disregard it and assume the total impedance (at AC) is dominated by our own provided termination network.

 

[andyr]

No problem

The filter designer designs filters which have equal input and output impedance. The speed controller PCB would need to have a 0.1ohm input impedance which seems very unlikely indeed. It'll be much higher than 0.1ohm in practice hence we can disregard it and assume the total impedance (at AC) is dominated by our own provided termination network.

Yes, I agree that Zin of the speed controller PCB must have a much higher Zin than 0.1ohm.

So are you saying that, because of this much higher Zin, we use a 0.1 ohm series resistor on the output of the pi filter, to provide the required termination. Then (from your post at 1:45pm) we follow that res with a series cap which is > 10 times the final pi filter cap value?

Now, if "the filter designer designs filters which have equal input and output impedance" then surely us putting 0.1 ohm on the output means we are assuming the Zout of the SMPSU (which is the input to the pi filter) is 0.1 ohm?


Thanks,

Andy

 

[abraxalito]

Yes to all the questions

Since 10,000uF at 50V is going to be quite bulky, I wonder what component values come out if we go for 0.2ohms? I guess it'll require 4,700uF which is rather more manageable. What's the current draw of your speed controller? I'd like to work out what's tolerable for voltage drop in the input termination.

 

[andyr]

Yes to all the questions

Since 10,000uF at 50V is going to be quite bulky, I wonder what component values come out if we go for 0.2ohms? I guess it'll require 4,700uF which is rather more manageable. What's the current draw of your speed controller? I'd like to work out what's tolerable for voltage drop in the input termination.

Aaah, OK - thanks, abraxalito.

Current draw of the speed controller is in the range 100-150ma.

If we take Zout as 0.2 ohm then:
* for a 7-element pi filter,
* for a 3KHz roll-off, and
* 0.1dB of ripple

... we get:
* L = 0.015mH / 0.017mH / 0.015mH, and
* C = 313uF / 556uF / 556uF / 313uF.

So we could indeed use a 4,700uF series cap after the 0.2 ohm series resistor.


Regards,

Andy

 

[abraxalito]

There's another aspect to the filters I haven't mentioned yet. The lower the ripple you go for, the shallower the slope. So I'd go for a minimum of 1dB ripple as in this applicaton the ripple (that's passband ripple but you only want to pass DC) is rather immaterial. With more ripple comes better HF attenuation.

With only 150mA current draw then even 0.5ohms is barely going to register so I'd go for this. Then your termination cap is smaller again. 10X was rather an approximation, going higher is better in terms of reducing any ringing.

 

[andyr]

There's another aspect to the filters I haven't mentioned yet. The lower the ripple you go for, the shallower the slope. So I'd go for a minimum of 1dB ripple as in this applicaton the ripple (that's passband ripple but you only want to pass DC) is rather immaterial. With more ripple comes better HF attenuation.

With only 150mA current draw then even 0.5ohms is barely going to register so I'd go for this. Then your termination cap is smaller again. 10X was rather an approximation, going higher is better in terms of reducing any ringing.

Aah, that is very interesting. Thanks.

0.5 ohms, 2dB of ripple and a 4KHz roll-off requires:
* L = 0.018mH / 0.019mH / 0.018mH, and
* C = 228uF / 308uF / 308uF / 228uF.

So I could use a 3,300uF series cap after the pi filter - am I correct? And I could use 220uF & 330uF caps in the pi-filter?

And the choice of 2dB of ripple should deliver quite steep slopes?

One point confuses me, though. If I have a series cap after the pi filter ... then surely this blocks DC? So how does the speed controller PCB - which is situated after the 3,300uF cap - draw the DC current it needs, to function?


Thanks,

Andy

 

[abraxalito]

The 'series' cap isn't as such. Its the termination, in conjunction with the 0.5ohm resistor. So the cap and the resistor are in series, but that series combination goes as an output 'load' of the filter.

Next up - do you really need a turnover frequency as high as 4kHz? The lower you go, the more attenuation you get higher up, because the roll-off starts earlier. Trouble is - lower frequencies need bigger caps, bigger chokes.....

You've already got very steep slopes - given that each element contributes 20dB/decade, with 7 elements that's 140dB/decade until parasitics come into play. If you want to see how it looks, I suggest downloading LTSpice and running a simulation.

 

[andyr]

The 'series' cap isn't as such. Its the termination, in conjunction with the 0.5ohm resistor. So the cap and the resistor are in series, but that series combination goes as an output 'load' of the filter.

Aah, gottit!! So the 3,300uF cap and 0.5ohm res are in series ... but they shunt to earth (from the output of the pi filter) - right?

Next up - do you really need a turnover frequency as high as 4kHz? The lower you go, the more attenuation you get higher up, because the roll-off starts earlier.

Well, yes, I could go down to 2KHz which, for the same parameters as before - namely, 0.5 ohms Zout and 2dB of ripple - delivers:
* L = 0.036mH / 0.038mH / 0.036mH, and
* C = 456uF / 617uF / 3617uF / 456uF.

So this would require a 4,700uF cap in the final 'shunt'.

But I thought the HF hash which the SMPSU is outputting (which we are using the pi filter to remove) would be above 100KHz - so the difference between a roll-off at 4KHz or 2Kz is moot? Particularly as you say I have an extremely high slope with the 7-component pi filter.

If you want to see how it looks, I suggest downloading LTSpice and running a simulation.

Yes, I have LTSpice ... and I know enough about it to create a circuit ... so I'll see how I go.


Thanks again,

Andy

 

[trobbins]

You have a turntable motor with a power source. You note that if the power source changes then the 'music' quality changes.

Perhaps before you spin off in one direction, it would be best to identify the causal connection between the two configuration, and then check that is correct.

I think you are inferring the connection is via 240VAC being used for powering each section of equipment, and some form of signal is ejected from the motor section, and makes its way conductively to the powering of the pickup amplifier and thence in to the music signal ?

Maybe the issue doesn't relate to conducted interference, but to emitted interference, or to mechanical vibrations via the platter caused by different motor drive current harmonics?

 

[andyr]

You have a turntable motor with a power source. You note that if the power source changes then the 'music' quality changes.

Aah, trobbins - you are in Melbourne. So am I. What a pity you couldn't have been at the listening sessions that were carried out over several weekends, with several different DC power supplies to the AC motor speed controller ... because then you wouldn't have had to ask these questions.

Yes, that is what we noted (the 'music' quality changes when the power source changes).

We tried - in order of increasing SQ result:
1. Meanwell SMPSU
2. 48v linear regulated (cathode follower) supply
3. #2 with an isolating transformer between it and the wall socket
4. #1 with an isolating transformer between it and the wall socket
5. 48v battery PS - which was 2 strings of 4x12v SLAs (2 parallel strings to lower the Zout of the PS).

So as the 4th option, above, is beaten by batteries, the speed controller designer suggested that the SMPSU must be injecting hash, along with the 48v supply - hence the search for a filter to chop this out.

Perhaps before you spin off in one direction, it would be best to identify the causal connection between the two configuration, and then check that is correct.

Weeelll - that's what we're trying to do.

I think you are inferring the connection is via 240VAC being used for powering each section of equipment, and some form of signal is ejected from the motor section, and makes its way conductively to the powering of the pickup amplifier and thence in to the music signal ?

No. With an isolating transformer in place - which should stop most (if not all) of the 'hash' produced by the SMPSU from polluting the phono stage and the preamp (which are also 'hidden' behind an isolating transformer), we reckon there must be hash being injected into the AC motor speed controller, which is interfering with its ability to drive the motor.

Maybe the issue doesn't relate to conducted interference, but to emitted interference, or to mechanical vibrations via the platter caused by different motor drive current harmonics?

Possibly. But I wouldn't think a battery supply has any "drive harmonics"? My aim is to build a LP filter for the 48v Meanwell SMPSU, to see whether this cuts out the HF hash and brings it up to the level of the 48v SLA supply. Because a permanent battery supply (rather than a cobbled-together experiment) will be expensive!


Regards,

Andy

 

[abraxalito]

Aah, gottit!! So the 3,300uF cap and 0.5ohm res are in series ... but they shunt to earth (from the output of the pi filter) - right?

Yep, spot on.

But I thought the HF hash which the SMPSU is outputting (which we are using the pi filter to remove) would be above 100KHz - so the difference between a roll-off at 4KHz or 2Kz is moot? Particularly as you say I have an extremely high slope with the 7-component pi filter.

It most likely starts a bit lower down, like around 40kHz. However with a 140dB slope for a decade then for an octave we'd get 42dB difference. Not to be sniffed at right? But in practice the response is going to be parasitics dominated - cap ESRs and layout quality will rule. So yeah the difference may well be moot, but due to implementation issues rather than due to the theoretical attenuation.

Bear in mind that by setting the roll-off to begin an octave sooner, you potentially save a couple of elements in the filter - you can use 5 components instead of 7, all things being equal (which they never are of course). Simulation with real-world ESRs/ESLs is going to be called for methinks to see if this saving can be realized in practice.

 

[DF96]

We tried - in order of increasing SQ result:
1. Meanwell SMPSU
2. 48v linear regulated (cathode follower) supply
3. #2 with an isolating transformer between it and the wall socket
4. #1 with an isolating transformer between it and the wall socket
5. 48v battery PS - which was 2 strings of 4x12v SLAs (2 parallel strings to lower the Zout of the PS).

These results suggest to me that if conducted noise is the issue, then the original SMPS PSU puts out less noise into its output than the linear supply or is more affected by incoming mains noise, so when isolated from the mains supply (assuming that is what the isolation transformer does, rather than merely having a safety role) it works better than the linear PSU. Maybe you should be concentrating on the mains side of the PSUs, not the output side?

I don't think you have anywhere near enough evidence that HF noise is the issue, unless you have further information you have not yet shared with us. The real issue could be hum or voltage variations or a design weakness in the speed controller.

 

[andyr]

Yep, spot on.

It most likely starts a bit lower down, like around 40kHz. However with a 140dB slope for a decade then for an octave we'd get 42dB difference. Not to be sniffed at right? But in practice the response is going to be parasitics dominated - cap ESRs and layout quality will rule. So yeah the difference may well be moot, but due to implementation issues rather than due to the theoretical attenuation.

Bear in mind that by setting the roll-off to begin an octave sooner, you potentially save a couple of elements in the filter - you can use 5 components instead of 7, all things being equal (which they never are of course). Simulation with real-world ESRs/ESLs is going to be called for methinks to see if this saving can be realized in practice.

Thanks, heaps.

OK, I will start the roll-off earlier ... but I will potentially still use 7 elements in the filter - it depends how the values turn out, from the calculator.

And I will do some simulations, this weekend.

Then we will have another listening session, to see whether this filter + the SMPSU "beats" the battery supply.


Thank you, muchly,

Andy

 

[andyr]

These results suggest to me that if conducted noise is the issue, then the original SMPS PSU puts out less noise into its output than the linear supply or is more affected by incoming mains noise, so when isolated from the mains supply (assuming that is what the isolation transformer does, rather than merely having a safety role) it works better than the linear PSU. Maybe you should be concentrating on the mains side of the PSUs, not the output side?

I don't think you have anywhere near enough evidence that HF noise is the issue, unless you have further information you have not yet shared with us. The real issue could be hum or voltage variations or a design weakness in the speed controller.

So why does the battery supply sound best of all (so far)?

Andy

 

[trobbins]

The source impedance of the supply powering the speed controller certainly would alter the harmonic current waveforms passing through the motor windings, and hence the mechanical torque variations induced in the platter. Discounting this mechanism is a little too early imho.

What you should have done in your testing is use a separate power supply to power the audio path (ie. battery powered, to eliminate conducted interference connection). If that had resulted in all tests 1-5 being the same in a double blind test then that would be conclusive.

The other issue is why is your audio amp system so susceptible to mains borne noise (presuming the conducted mechanism pathway). Did you try some other better designed gear?

Ciao, Tim