CLCLC filters in BeePre and Kaiju

[Deke609]

Last night I modified the stock CRC filter in my Beepre to a CLCLC filter: 220uF -> 5H 105 DCR -> 220uF -> 1.5H 56 DCR -> 440uF.  I made this change "just to see". Annotated photo and PSUD sims of stock and modified power supply attached.  In simulation at least, the new combination matches the stock output voltage closely, reduces ripple to almost nothing, and adds a total of 440 uF of extra capacitance (which wasn't a goal - it just worked best this way in sims).

My BeePre was already modded before I made last night's change: (1) 1/2W 110R Vishay Dale grid stoppers on the 300Bs; and (2) the capacitance of the cathode resistor bypass caps has been increased from 10k uF to 32k uF (the big caps in the center of the amp are 22k uF 80 VDC caps - I've posted aobut this in a different thread).

I previously made a similar change to the power supply filter in my Kaiju - where I thought the CLCLC filter made an audible improvement, but only a very small one (in comparison with which using film caps as the last filter caps in the Kaiju produced a tiny and quite possibly imaginary change; and changing the lytic cathode resistor bypass cap to a film cap resulted in a much bigger improvement). So I wasn't expecting much if any of a change in adding a CLCLC filter to the Beepre, but the result surprised me: the BeePre sounds very noticeably better. it is at once more detailed and smoother, and with "energetic" music, it is faster - so much so that it sounds like I'm listening to a completely different performance of the same music -- just two examples: "Whipping post" by the Allman Brothers, and pretty much every song on Paul Simon's Graceland album. Whereas "speed" seems unaffected with other music -- e.g., Amber Rubarth's acoustic Sessions from the 17th Ward -- and all I hear is added detail and ease.

The above listening impressions are inconsistent with those of at least Aaron J (I will post a link to his impressions in a later edit), who IIRC replaced the stock CRC filter with a CLC filter (I think a 10H). I'm not sure what to make of this. My PSUD sims suggest that two LC filter stages, even with lower total inductance than one might have with a single stage, reduces ripple to a greater extent than a single stage.  And perhaps the added capacitance plays a role.  Or it's just my imagination - but I don't think so: the experience of listening to a piece of music with which I am very familiar and having it sound not just "better" but like a completely different performance suggests to me that the change is real. I had a similar experience with my first Stereomour II after adding the Shunt Reg upgrade.  The sonic effects of the CLCLC change isn't of the same magnitude as the SR Stereomour upgrade, but it's in the same ballpark (to my ears). Now, this may seem crazy, since both the BeePre and Kaiju are already regulated out the wazoo. How can adding some PSRR further up the power supply chain make any difference when ripple is already reduced further down the chain to negligible levels? I dunno. Maybe it's just the added capacitance. Or maybe it still makes a difference. Regarding the latter possibility, I am reminded of some of PJ's comments on the AA forum (See my "Couple of Quotes ..." post in General Discussion some months ago) to the effect that he had been experimenting with stacking CCS and shunt reg isolation from the power supply and could hear a difference even when accepted engineering standards suggested that the changes would be inaudible.

Now for my reasons for titling this thread "CLCLC filters in Beepre and Kaiju": I suspect (but do not know) that part of what I am hearing is the result of the CLCLC change to the Kaiju, and that the changes to the power supplies in both BeePre and Kaiju are interrelated and additive. My hypothesis is that they are each part of one larger PSRR change. My thinking is as follows: if residual 120 Hz ripple in the BeePre adds some noise, that noise becomes part of the signal fed to the Kaiju and will get amplified, regardless of whether the Kaiju has a less-noisy CLCLC filter. In which case the CLCLC mod to the Kaiju may not seem to do much. But reducing the 120 Hz related noise in the BeepPre with a CLCLC filter allows one to hear the cumulative effect of both - anyway, that's my mostly uneducated guess about what might be happening here.

Some installation notes. The 1.5H chokes are only 1.5 inches wide, and so fit fairly easily alongside the pcb-covered power transformers. The 5H chokes, however, are 2 inches wide and 3.25 inches long. One fits diagonally near the IEC inlet, but to fit the other I had to remove one pair of the output jacks and one pair of the input jacks - so now I have 4 big holes at the rear of the chassis that I will probably cover with some mesh screen.  So the aesthetics took a bit of hit - but I'm ok with that. Other may not be. But I think this could be avoided by mounting the chokes on a t-shaped bracket so that the chokes stand parallel to the chassis rather than perpendicular to it. Or maybe you could mount them on top of the chassis.

I should note, too, that in installing the chokes in both BeePre and Kaiju, I added a "flux band" to each choke. The band wraps around the outside of the choke in parallel with the direction of winding (so around both the winding, the EI core and the frame). The bands are three-layers: 30 gauge copper sheet, some EMI shielding foil called MCF5, and a very thin layer of copper foil tape around the two previous layers to hold them together. The band is wired to chassis ground.  The MCF5 EMI foil is made by a German company that claims a -30dB attenuation of low frequency interference (IIRC this encompasses all frequencies in the audible bandwidth) and much greater attenuation of high frequency interference. I obviously cannot confirm this. I can post a link to the datasheet if anyone is interested (although it is easy to find with a google search).

Lastly: a product recommendation. In the photo you can see some clear thick tape wrapped around the big 22k uF Nichicon caps. It is silicone gel tape that you can find on Ama*zon and other online retailers.  It has some very useful properties for affixing and separating parts in an amp: it is sticky and strong as hell, but can be peeled off (with effort - this stuff really grabs) without leaving any residue; it is electrically insulating; and presumably b/c of it's gel-like consistency adds some vibration damping.  It is a bit difficult to cut, though. I recommend a strong and sharp pair of utility or kitchen shears.

cheers, Derek

[Paul Birkeland]

The hybrid shunt regulator in the BeePre takes care of any remaining power supply ripple.  The noise floor in the BeePre is dominated by the filament power supply, not the B+ supply, so I would not come to any conclusions about lower noise floor being responsible for anything you're hearing, as the noise floor is going to look the same with the two different power supplies.  The B+ rail is already quiet enough that I can't even resolve much of anything that looks like audio frequency noise with my scope, let alone my ears.

It's not all that surprising that things sound a little different with different parts installed.  I always find it's most helpful to revert back to what you had before in 3-4 months to verify the results.

[Deke609]

I always find it's most helpful to revert back to what you had before in 3-4 months to verify the results.

Good advice. I may do so over the winter. My preference would be to leave the last 220uF additions that are in parallel with the last stock 220uF and just disconnect the chokes and re-insert the 300R 10W resistor - only b/c I'm lazy and don't want to remove those giant Nichicon caps again. And installing the new 220uF caps in parallel with the stock was a bit tricky - not a lot of space for fingers or even slim profile pliers.  But whether I can leave them will depend on whether PSUD shows that the stock CRC will work just fine with the added capacitance.

cheers and thanks, Derek

[Paul Birkeland]

If you want to look at what the regulator is doing, put a node in PSUD that's a 1000H/600 ohm choke and about a 10uf cap.  Since PSUD won't let you do 1000H, you could approximate it as CLCLCLCLC where each L is 250H/150R and all the middle caps are 0.001uF.

[Deke609]

@PB - I can't get the PSUD sims of the regulator to work. What I tried: adding 4 x LC stages of 250H/150R and 0.001uF cap, with and without the last cap as a 10 uF. I tried tacking this on to both the stock filter and my CLCLC filter. With the 10uF as a the last cap, I get some sort of timeout error and the sim stops. Without the 10 uF cap at the end, I get totally screwy results. See attached screenshots.

Is my setup wrong? When you say "add a node", is that some sort of special configuration in PSUD? THe only other option I see besudes adding a regular filter stage or load is a "current tap". Is that what I use?

many thanks, Derek

[Paul Birkeland]

We can be very thankful that the C4S doesn't ring!

[Deke609]

It would be very cool to be able to model the C4S and shunt reg to see if there's any apparent reason for what I'm hearing. Maybe I need to bite the bullet and learn how to use LTSpice.

I did run a PSUD sim of the stock filter with 440 uF instead of 220 uF and it works fine and looks pretty close to the original. So when I get up the gumption to remove-then-reinstall those big nichicon cathode resistor bypass caps, I'll try reverting to the (mostly) stock CRC filter (but with 440 uF at the end). 

cheers, Derek

[Deke609]

@PJ or @PB - I'm still interested in exploring this, even if only theoretically and not empirically, b/c I swear I hear a significant difference (and improvement) with the CLCLC filter.

In my thread about a making a DIY mains voltage quasi-regulator, PJ advised that my proposed diode voltage trimmers would cut the peaks off the sine wave and produce spikes at odd multiples of the base frequency. That got me reading about diode rectifiers and harmonics. From what I've read it appears that any full wave diode-based rectifier produces odd-order harmonics - and that one of the ways of  mitigating this is to use filter chokes.

So this has me wondering: if the ripple reduction achieved by the CLCLC filter is negligible compared to the reduction achieved by the C4S, could reduction of harmonics in the power supply have an audible effect? 

Any thoughts?

cheers and many thanks, Derek

[Paul Joppa]

Your diode voltage trimmers would cause an anomaly at the zero-crossings, not at the peaks. The regular rectifier causes a small squishing of the sine wave at the peaks. And diode switching can excite the power transformer winding resonance, especially if it's a silicon diode which also has reverse-recovery spikes. That's four different effects.

Just wanted to clarify - things are always more complicated than you think at first!

[Deke609]

Many thanks PJ.

That's four different effects. ... things are always more complicated than you think at first!

Yeah, no kidding! Well, that gives me 4 different things to read up on. I think I'll start with the possibility of transformer ringing. One follow-up question: do you think that power supply chokes could damp transformer ringing?  Just by happenstance, I'll be playing around with transformer snubbering sometime in the next couple of week using a "no math" snubber calculation pcb (for those interested in this, see here: https://www.diyaudio.com/forums/power-supplies/243100-simple-math-transformer-snubber-using-quasimodo-test-jig.html)

I'd really like to figure out, if it all possible, why I hear what I hear. What really puzzles me is why the same mod (a CLCLC filter) would, to my ears/brain,  make at most a small (and admittedly quite possibly imaginary) difference in the Kaiju, but a big difference in the BeePre (where I have a much harder time allowing for the possibility that it's imaginary).

cheers and thanks, Derek

[Jamier]

Derek,

     Google: snubbers and rectifier ringing Morgan Jones. You’ll see a PDF associated w/ desmith.net.
I hope you find it helpful.

Jamie

[Deke609]

Many many thanks Jamie - both interesting and helpful. I was 80% through the article and thinking "Well dammit to hell, MJ says I need to be snubbering the diodes themselves, not the transformer secondary" but then got to the end where he writes that a snubber across the secondary is the way to go - good! b/c that's in line with the approach Mark Johnson's "no math snubber pcb" takes:  figure out what frequency will cause the worst transformer ringing and then add a snubber to damp it.  Although the possibility of ringing at multiple different Khz/MHz frequencies is a bit of a concern. But in such a case, at least I should be able to tame the worst offender.

cheers and thanks, Derek