Go Back   Home > Forums > Blogs > abraxalito

Rate this Entry

Amplifier-centric two-way XO (and system) design

Posted 3rd July 2014 at 04:30 AM by abraxalito
Updated 5th July 2014 at 02:55 PM by abraxalito

The traditional view is that the speakers are the weakest link of the audio system and hence they're not really designed with amplifiers in mind. Why make amp design easier when speakers are plenty hard enough to get right?

In my experience speakers don't contribute very much to the listening satisfaction. They're more a matter of taste, but a much better speaker can't undo the faults of an amplifier (or DAC for that matter). That's not to say some speakers don't sound better than others - I mostly prefer my Paiyons to the much cheaper Denons but their 'betterness' is orthogonal to the way to the manner DACs and amps sound better. Electronics has had the biggest impact on my listening satisfaction so far so here is an XO designed to make the electronics' job as easy as possible.

Given that the critical parameters of an amp designed for listener satisfaction aren't THD+N and FR, rather multitone IMD (aka MTPR) and PSRR it makes sense that an amp is given the narrowest possible frequency band to handle. With fewer distinct frequencies present there is going to be lower IMD. Similarly PSRR of an amp is a function of decoupling efficiency and tts much harder to decouple an amp over a wide frequency range than a small one, so narrow band amps are preferred for this reason too.

All this assumes though that an XO can split the frequency bands without introducing IMD or PSRR issues of its own, fortunately passive XOs have this ability needing no power supply and producing vanishingly low IMD at low levels.

In this case I've chosen the TDA8932 as the bass/mid amp and that struggles to deliver clean sound in the HF due to limited PSRR at higher freqs. Yet its bass is the best yet, so it is really suitable for bass/mid duty provided the bandwidth its being asked to deliver is kept as narrow as possible. Hence here I've chosen a steep slope LPF to minimize the IMD generated in the amp and ease its decoupling burden. Time domain performance isn't a concern at this stage, nor is speaker directivity control. I'm curious to see how the amp performs when given the narrowest band of frequencies to reproduce.

The tweeter is being driven via the same type of tweeter bandpass filter I designed last month but with a couple of tweaks. I've moved the crossover frequency lower to 2.5kHz as when I put the tweeter on the siggen, it sounded fine here and looked to have a resonance around 2kHz by which time the bandpass filer is -26dB. The lower the XO frequency the better here from the point of view of the TDA8932 - also it means this XO should be at least suitable for a 6.5" bass/mid eventually. I also decided to try out the T-network instead of the pi-network topology as in this application I do like to have an input inductor and AC couple with capacitor anyway. So why not save on a couple of parts?

The post amplifiers will be AD815s as usual but I've found that a 2nd order filter (single ended) between these and their output trafos helps lower the perceived noise even more..

Update : It has occurred to me that Lynn Olson's thread 'Beyond Ariel' is another example of amplifier-centric system design. For Lynn though the constraints are rather different - he's not inclined to increase the number of amplifier channels (since his all-triode Karna monoblocs are rather expensive to build) so is using passive XOs. For me I consider that decent sounding (solid state) amplifier watts are going to be considerably more affordable than that, hence I'll just go with as many channels as called for to obtain the best sound.
Attached Thumbnails
Click image for larger version

Name:	twowaysch.jpg
Views:	261
Size:	56.4 KB
ID:	1319   Click image for larger version

Name:	twowayXO.jpg
Views:	150
Size:	83.3 KB
ID:	1320  
Posted in Uncategorized
Views 941 Comments 5
Total Comments 5


  1. Old Comment
    wintermute's Avatar
    Hi Richard, are you putting realistic parasitics into your coils and caps in spice? I found when simming passive filters that the performance degraded markedly when realistic small amounts of esr and esl were added to the caps, and dcr for the coils.

    Posted 3rd July 2014 at 08:11 AM by wintermute wintermute is offline
  2. Old Comment
    abraxalito's Avatar
    Hi Tony - yes the R is the killer of sharp corners for passive filters. Fortunately I have a supply of gapped ferrite pot cores which allow me to achieve Qs into 3 figures at 20kHz. Getting such high Qs at 3kHz is more of a challenge but can be done by going to slightly larger coils. I've not had to use Litz yet but I'm getting close....
    What kinds of inductors are you using? I've not found any commerical ones that beat hand-wound pot cores for low AC losses. ESRs are an issue with X5R and X7R ceramics - I've gone over to using NP0 for this filter to get lower ESR. ESL is never a problem for me with SMT caps in audio signal applications, it becomes a series issue in decoupling though.
    Posted 3rd July 2014 at 01:21 PM by abraxalito abraxalito is offline
    Updated 3rd July 2014 at 01:24 PM by abraxalito
  3. Old Comment
    wintermute's Avatar
    Hi Richard, I was only simming passive to get a baseline for my active implementation of the passive filter. I wound my own air cored inductors for my passive crossovers but they are speaker level not line level.

    I know planet10 has commented a number of times about pllxo's getting "droopy" when cascading more than one order, and my sims certainly confirmed that.

    I wasn't using smd so was putting in 6nH for esl for caps (which is the spec of the rifa MKP1387's I'm using. for coils even 0.1 ohms is enough to cause some serious deviation from ideal, I was quite surprised.... In my case though I was simming with 1 ohm load since that is the "normalised" filter for doing the active calculations... perhaps I should have used a more realistic load and then scaled the component values appropriately....

    Posted 4th July 2014 at 08:26 AM by wintermute wintermute is offline
  4. Old Comment
    abraxalito's Avatar
    When planet10 and I got into a conversation about PLLXOs some time ago he wasn't hungry enough (it seemed to me) to be prepared to make them work into a low enough impedance. I got the distinct impression he wanted a PLLXO that worked into 100k load. In which case, sure he'd have desperate problems with getting high enough Q. If your load is just 1ohm, then 0.1R is about 1dB droop already. I have a filter on my blog which has a 0.93ohm load and doesn't suffer droop but its made from custom ferrite bead inductors. As a general rule of thumb for not-too-demanding filters the total DCR of all the series inductors shouldn't exceed 15% of the load impedance. My AIF with 4 inductors around 1.6R each falls within this guideline for example with a 50R load. It does still exhibit about 1dB of droop but seeing as the zero-order hold of the DAC causes droop beyond this, there's already a network in place to correct for it.
    Posted 4th July 2014 at 11:20 AM by abraxalito abraxalito is offline
  5. Old Comment
    wintermute's Avatar
    Thanks, that confirms what I was thinking. I shall scale and do some sims with a more friendly load

    Posted 7th July 2014 at 08:11 AM by wintermute wintermute is offline

New To Site? Need Help?
Copyright 1999-2017 diyAudio