Back to passive filters
Posted 8th May 2013 at 02:11 AM by abraxalito
I didn't much care for the sound of my active elliptic filter - great dynamics in the bass for sure but the upper-end colourations were a bit unnatural sounding. So I've shelved tthat one for now and instead I'm playing with a simplified (by which I mean fewer inductors) passive elliptic.
There are two topologies for building elliptics where the zeroes are realized either by shunt series-LC networks or series paralleled-LC networks. The series created zeroes means fewer inductors are called for. In its most basic, unbalanced form there would be just three inductors for a 7th order filter. This filter though is balanced and designed to feed my Nitro desktop amp directly, without any I/V amplifier stage.
There are two topologies for building elliptics where the zeroes are realized either by shunt series-LC networks or series paralleled-LC networks. The series created zeroes means fewer inductors are called for. In its most basic, unbalanced form there would be just three inductors for a 7th order filter. This filter though is balanced and designed to feed my Nitro desktop amp directly, without any I/V amplifier stage.
Total Comments 13
Comments
-
Interesting. Could you be more specific about precisely what those "upper-end colourations" were? This is the area where one is more and more balancing on the knife edge of "correct" sound, and the slightest weakness anywhere in the whole chain can be severely emphasised ..
From personal experience, the best sound is often directly adjacent to some of the worst sound: 1' of an arc to the left, unlistenable; 1' of an arc to the right, tediously dull; inbetween, just right ...Posted 8th May 2013 at 03:42 AM by fas42
-
No I am not going to be much more specific in this instance as I didn't listen intently enough to understand what the colourations were. I just didn't enjoy listening to this circuit as much as the earlier passive one. But I can give you a hint from the kinds of music I listened to - on piano at the top end of its register, some notes sounded too right to be true (like hyper-real) and others missed the mark. Leonard Cohen's voice wasn't as sexy, brass lost some of its intense bite and became more electronic--y. This could easily be caused by having resonances in the power supply impedance at particular frequencies but its just speculation.
The original reason I did the active filter was to get a more portable (smaller size primarily) solution but when I'd optimized the power supplies with enough caps for low impedance at HF I realized there was practically no size advantage after all. So passive wins on all fronts but ease of manufacture. Probably I'll find someone to wind the coils with sufficient precision so even this will get sorted...Posted 8th May 2013 at 05:13 AM by abraxalito
Updated 8th May 2013 at 05:28 AM by abraxalito -
There's a small twist to the story of the poor sounding active filter - it was inadvertently built with noisy shunt regs. I have found the resistors I used have oodles of excess noise (0603s really aren't recommended unless the high stability thin-film types). So I'm modding the shunts to take thin-film (I happen to have a couple of reels) and for the other voltage setting resistor whose value I don't have in thin-film, I'm substituting a 1206 type thick film which are more benign. Then I'll have another listen....Posted 23rd May 2013 at 02:35 PM by abraxalito
-
A bit of the ol' [I]everything matters[/I], yet again ...Posted 24th May 2013 at 01:18 AM by fas42
-
Yes everything matters, but not everything matters equally. I had the same 0603 resistors for I/V duty and figured if I could hear the noise (and I could with these shunts in the DAC PSU) then maybe I could hear it in the I/V. So I swapped them out for thin-film. Can't say I've noticed a difference. Whereas 'received wisdom' is that I/V resistors are inordinately sensitive to construction quality. Perhaps my system is just not sufficiently resolving
Posted 24th May 2013 at 03:49 AM by abraxalito
-
Got to knock off the big'uns before the littl'uns show their faces, :). This is a round and round and round game, it's easy to ditch, dismiss a tweak at one stage because its impact is swamped by a "bigger" problem ... only to have it become crucial when the system is more "resolving", :D.
"Vicious" recordings are a good way to spot whether there is some subtle interaction going on, which is not obvious with cleaner material ...Posted 24th May 2013 at 04:43 AM by fas42
-
Yes - it seems to me that its all a question of 'what's the low hanging fruit?'. If you listen to people like CopperTop (oops, sorry Groucho) or Tim on WBF, seems they consider the low hanging fruit to be passive speakers. Make 'em active first they say! I used to hold that view, but my current journey has found the lowest hanging fruit to be the DAC. Then the amp. Only now I've found out how hard it is to get a decent PSU for an amp do I consider going active. That's for the amp's benefit though, which is not a mainstream reason as far as I can see - most will say its because passive XOs are the lowest hanging fruit.
Having said that I realize on reflection that I haven't got the timeline quite right. Looking back on this blog, I started out with active speakers and found poor implementation - particularly grounding. Then I went on to power supply filtering. Then I worked on DACs, still feeding my active speakers. I only went back to passive speakers when I wanted to develop amps because of not wanting to build more than two channels at a time for SQ evaluation and also not to rely on questionable transparency active XOs. It was only then that I discovered that passive speakers did not suck as badly as I'd previously thought.
Incientally on amps, today I've made some progress on simulating PSRR in LTSpice, something we touched on when comparing chipamps. I have a discrete schematic of LM1876 which matches the OL gain and PSRR in the DS for the LM1876. Its based on the schematic shown in the DS, with some minor tweaks. For example I need 20pF comp capacitance rather than the 10pF shown. There's also a transistor upside down in the first current mirror. What is most interesting is - I do not get PSRR as high as shown in the DS when I have a closed loop amp. It follows the HF plot very closely but the PSRR referred to the output tops out at 51dB below 10khz and stays stuck there as the freq goes down. Wanna play with the schematic?
Maybe won't be necessary - I've found an unbalanced current mirror is at least partially responsible for this, work in progress...Posted 24th May 2013 at 07:36 AM by abraxalito
Updated 24th May 2013 at 08:26 AM by abraxalito -
Yep, sounds interesting ... send it along when you reckon you've got it sorted ...
Thanks,
FrankPosted 24th May 2013 at 10:26 AM by fas42
-
OK I am gradually getting the hang of understanding PSRR better, let's see how my explanation goes.
After fixing up the imbalanced current mirror (two resistors in the emitterss - one got changed at some point) now the PSRR in the normal 26dB closed loop gain maxes out at 65dB below 1kHz. This was the anomaly I had been scratching my head over for a few hours. That's because I expected to see it climb as shown in the DS PSRR graph, reaching 103dB minus the closed loop gain of 26dB or about 77dB. But in fact its 12dB short of this.
My understanding of how this discrepancy arises is that those PSRR plots are for the open loop case and are referred to the input. In such a case the input signal is obviously very small, there's very high OL gain at LF. So what dominates the PSRR is the supply rejection of the input stage. This is rather akin to noise figures being dominated by input stage noise in high gain configurations.
However in the CL case with only 26dB of gain the output signal is much smaller and so tthe output stage PSRR can dominate the figure - which is what I think is happening. When feedback is applied the limiting factor is no longer the input stage's PSRR rather the output stage's. The figure of 65dB seems remarkably similar to that achieved by the LME49600 buffer. Obviously that one is output stage limited as its only an output stage
So if this understanding is correct then the usual chip amps do not in fact offer better PSRR at lower freqs (than the TDA8566) as I had previously imagined going by their PSRR plots. Also what follows from this (if its correct) is there's precious little point in filtering the small signal stages (as can be done on the TDA7294 for example) because the damage is being done at the output stage. So this would mean chipamps do not in fact lose out to discrete amps where the small signal stages can be filered to an arbitrarily high level - both are in fact output stage limited in practice.Posted 24th May 2013 at 04:38 PM by abraxalito
Updated 24th May 2013 at 04:52 PM by abraxalito -
Pressing on with the sim to test my hypothesis - so far all results are from running positive rail PSRR. This is the better one, when I've understood this I'll move on to the negative rail.
Installing a hefty RC on the positive rail which only leaves the two output emitter follower stages (driver & output) exposed to the ripple gives 110dB below 80Hz. This seems too good - I rather suspect the models now, particularly for the output devices. Perhaps these transistors have no output (collector-base) capacitance?
So it looks as if it might not be the output stage, rather the TIS (or VAS) which is the culprit. However protecting only this stage from the ripple while having the input stage exposed gives a very poor plot (power supply gain above a few kHz) so this approach looks invalid. Anyway going back to the detail of the plot - with only the driver and output stages seeing ripple, we have about 61dB PSRR at 20kHz, from the +ve rail. Moving the driver collector to the clean supply, the performance gets amazing (too amazing so the models must be suspect) - 126dB below 1kHz and 107dB at 20kHz.
So rather looks like my 'output stage is the culprit' hypothesis has taken a fatal hit. That is until I verify what the output transistor models really are. Stay tuned.....Posted 25th May 2013 at 12:21 AM by abraxalito
Updated 25th May 2013 at 12:41 AM by abraxalito -
Sounds, as in fact is obviously the case, that getting the accuracy in the model is crucial. Real tests on real sample, [I]and [/I]using extremely tightly controlled behaviours of the driving voltage sources, and also done at various loads, will be the only way to get the full story ...Posted 26th May 2013 at 12:31 AM by fas42
-
Here's another interesting conundrum - compare and contrast the LM1876 against the LM4766. The schematic has changed in a couple of fairly subtle ways - the upside down current mirror transistor has been corrected, but the other change I've so far noticed is the introduction of an error - a wire crossing at the driver stage is now a connection. I've simmed both and think the LM1876 schematic is correct here. Oh and the output ballast resistors have gone down in value. Nothing else I can see has changed.
But then the positive rail PSRR plots are amazingly different - so what can possibly account for this?Posted 26th May 2013 at 01:08 AM by abraxalito
-
Because they slightly varied the test and measurement conditions between the two? Note that the LM1876 graph seems to be a bit more 'realistic', the LM4766 has been 'cleaned up'.
And, different samples? In the characteristics table they're terribly pessimistic about how poor an actual sample can be, a difference of 40dB is allowable -- would be quite easy to get a real "dud"! What could be so variable in the manufacture to make such a difference ...?!Posted 26th May 2013 at 10:58 PM by fas42
