Thank you enormously for taking the time to do all that, I very much appreciate it!!
I didn't actually know what the max grid/mains voltage could be physically allowed to reach, so thats most interesting. Previously I've gone by rules of thumb for normal operating voltage tolerances, but i feel happier working to actual limits in case of abnormal situations. I'd imagine that the capacitors also have a margin for error over and above their rating, particularly for short term spikes, but I prefer to see such things as safety margins rather than something I'd intentionally plan to use.
Thank you for the plot too, thats extremely kind; I'm very much a noob when it comes to such calculations and simulations. "..a trapezoidal numerical integration of the crossover transfer function times the spectral density plus some normalization" is unfortunately way over my head, but thanks to your plot I understand whats going on. Your assumptions on the closed box are likely pretty close and I will be playing a wide range of material from folk and unaccompanied vocals to rock and even movies, so it would be unsafe to bias my design towards any one genre. It is very interesting to see that the bass and tweeter actually cross over towards the RHS of the plot; I would have expected that sort of trend but had no idea it would be of such large magnitude.
Perhaps i will reconsider the allocation of capacitors, then. TBH I was already nervous about such a small reservoir on the tweeter amp supply (for no good reason, but purely subjectively it seemed too small to have much leeway), and the mid range is most important to me so I don't want that to suffer either.
Though I suppose its all relative and I should keep things in perspective once I get into the stage of diminishing returns; the main reservoir of my current marantz integrated amp comprises only a single 10,000uf cap per rail, and thats for full range on both channels. Part of the joy of going the DIY route is that its easy to improve on things like that, but I am liable to get carried away if I'm not careful.
Cheers
Kev
I didn't actually know what the max grid/mains voltage could be physically allowed to reach, so thats most interesting. Previously I've gone by rules of thumb for normal operating voltage tolerances, but i feel happier working to actual limits in case of abnormal situations. I'd imagine that the capacitors also have a margin for error over and above their rating, particularly for short term spikes, but I prefer to see such things as safety margins rather than something I'd intentionally plan to use.
Thank you for the plot too, thats extremely kind; I'm very much a noob when it comes to such calculations and simulations. "..a trapezoidal numerical integration of the crossover transfer function times the spectral density plus some normalization" is unfortunately way over my head, but thanks to your plot I understand whats going on. Your assumptions on the closed box are likely pretty close and I will be playing a wide range of material from folk and unaccompanied vocals to rock and even movies, so it would be unsafe to bias my design towards any one genre. It is very interesting to see that the bass and tweeter actually cross over towards the RHS of the plot; I would have expected that sort of trend but had no idea it would be of such large magnitude.
Perhaps i will reconsider the allocation of capacitors, then. TBH I was already nervous about such a small reservoir on the tweeter amp supply (for no good reason, but purely subjectively it seemed too small to have much leeway), and the mid range is most important to me so I don't want that to suffer either.
Though I suppose its all relative and I should keep things in perspective once I get into the stage of diminishing returns; the main reservoir of my current marantz integrated amp comprises only a single 10,000uf cap per rail, and thats for full range on both channels. Part of the joy of going the DIY route is that its easy to improve on things like that, but I am liable to get carried away if I'm not careful.
Cheers
Kev
Last edited:
Perhaps it may also be worth me considering a middle-way of having the caps shared by all three amps, but to still split the shared bank with resistors. This would retain the CRC configuration that generally seems to find favour for smoothing, but without the complexity of different secondary stages per amp. There seem to be a range of challenges, decisions and concerns with trying to divide up the second stage into three, and I have the potential to make it worse rather than better, whilst a shared one would allow pooling of resources to suit all genres of music and avoid the uncertainties and difficulties.
So I think I'll 'very' seriously consider re-allocating my 4,700uf caps into banks of five as: 23,500uf -R- 23,500uf per rail. Surely 46,000uf per rail (92,000uf per channel) can't be bad... across both left and right speakers thats not far off 200,000uf, which is 10x what my current hi-fi amp has.
The 2200uf audio-grade caps I'd intended for the treble amp could then be reallocated as local/bypass caps right up at the amps themselves, which seems like a suitable use for them.
Cheers
Kev
Last edited:
SMPS ?
I would go for an SMPS like the Connexelectronic and an extra 10 mf cap on each rail .
protected , stabilized (no rail sag at high volume ) , very reliable , low ripple and cheap .
Cheers,
Rens
I would go for an SMPS like the Connexelectronic and an extra 10 mf cap on each rail .
protected , stabilized (no rail sag at high volume ) , very reliable , low ripple and cheap .
Cheers,
Rens
Last edited:
Yeah, 300-600W is a natural SMPS application. The tradeoffs are mainly reduced thermal requirements, potentially increased SFDR due to reduced ripple, and potentially decreased SFDR due to the ripple being at higher frequencies where PSRR is lower. Reuse of existing project iron aside, I'd say the design advantage of a linear supply is mainly it's easier to figure out how the amplifiers and supply interact. Minimizing this interaction by throwing capacitance at the problem is certainly valid, though it gets kinda big, expensive, and generates additional complexity in the form of soft start requirements.
To revisit post 26, what the 92,000uF idea yields is something around a 15dB reduction in ripple over the more 14,000uF-ish approaches. This feeds through into a 15dB increase in SFDR with respect to the supply. Alternatively, if one leaves the amount of capacitance the same but increases the amplifiers' PSRR by 15dB then SFDR also increases by 15dB. Project 101 is not specified for PSRR but discrete amplifiers usually have around 70dB. With care in design it's possible to increase this to 80 or 90dB but it's extremely difficult to get sufficiently good matching to go above that in a discrete build. In comparison, integrated solutions are usually smaller, cheaper, easier to implement, and the manufacturer takes care of the matching. The LME49830 is the most commonly used device for FET builds in this power range, though the LME49811 can work as well. Both have 105dB PSRR typ. So one can sweat the supply as we're doing here to get improvements in the 10 to 15dB range. Or one can get 35dB by making the amp channels easier to build. I don't mean to be a downer but, apropos of this topic, it'd be poor of me not to also point out there's not exactly a shortage of 70 cent regulators which offer 60 to 80dB PSRR.
As used in the majority of linear audio supplies CRC and CLC are mainly a class A solution as they lack load regulation and therefore handle class B/C/D/G/H/T... poorly (when the load bandwidth is higher their cutoff). Class A load obviates the need for load regulation but passive filters' limited line regulation can still be an issue. There are applications where pi filters cornering at, oh, 10kHz and higher, are superb. But for supply management below a few kHz linear and switchmode regulators generally offer better performance with lower cost and implementation complexity. Capacitance multipliers provide a middle ground which you may find interesting to check out as well.
You're welcome for the analysis. It happens I like sweating supplies so I'm pleased it was useful to you.
Some classes of capacitors are specified for surge voltage but the electrolytics applicable to typical audio supplies aren't. So designs which attempt to take advantage of undocumented margin are relying on behaviour not guaranteed by the manufacturer. I've encountered devices which do this and have received regulatory certification but, at best, it's shoddy engineering. DIY lacks commercial pressures so I can't see any reason to take the risk.
To revisit post 26, what the 92,000uF idea yields is something around a 15dB reduction in ripple over the more 14,000uF-ish approaches. This feeds through into a 15dB increase in SFDR with respect to the supply. Alternatively, if one leaves the amount of capacitance the same but increases the amplifiers' PSRR by 15dB then SFDR also increases by 15dB. Project 101 is not specified for PSRR but discrete amplifiers usually have around 70dB. With care in design it's possible to increase this to 80 or 90dB but it's extremely difficult to get sufficiently good matching to go above that in a discrete build. In comparison, integrated solutions are usually smaller, cheaper, easier to implement, and the manufacturer takes care of the matching. The LME49830 is the most commonly used device for FET builds in this power range, though the LME49811 can work as well. Both have 105dB PSRR typ. So one can sweat the supply as we're doing here to get improvements in the 10 to 15dB range. Or one can get 35dB by making the amp channels easier to build. I don't mean to be a downer but, apropos of this topic, it'd be poor of me not to also point out there's not exactly a shortage of 70 cent regulators which offer 60 to 80dB PSRR.
As used in the majority of linear audio supplies CRC and CLC are mainly a class A solution as they lack load regulation and therefore handle class B/C/D/G/H/T... poorly (when the load bandwidth is higher their cutoff). Class A load obviates the need for load regulation but passive filters' limited line regulation can still be an issue. There are applications where pi filters cornering at, oh, 10kHz and higher, are superb. But for supply management below a few kHz linear and switchmode regulators generally offer better performance with lower cost and implementation complexity. Capacitance multipliers provide a middle ground which you may find interesting to check out as well.
You're welcome for the analysis. It happens I like sweating supplies so I'm pleased it was useful to you.
Some classes of capacitors are specified for surge voltage but the electrolytics applicable to typical audio supplies aren't. So designs which attempt to take advantage of undocumented margin are relying on behaviour not guaranteed by the manufacturer. I've encountered devices which do this and have received regulatory certification but, at best, it's shoddy engineering. DIY lacks commercial pressures so I can't see any reason to take the risk.Kev,
I really think you are over-thinking this. Unless you are planning on providing the speaker stack for a Deep Purple concert, a simplified power supply with a single bank of caps per rail will be fine. As I said earlier, I power my active speakers, which use four LM3886 amps, from one 160VA transformer and three 4,700uF caps per rail.
The rails may sag under load but, by then, the load is loud enough to force some people to cover their ears.
Abs
I really think you are over-thinking this. Unless you are planning on providing the speaker stack for a Deep Purple concert, a simplified power supply with a single bank of caps per rail will be fine. As I said earlier, I power my active speakers, which use four LM3886 amps, from one 160VA transformer and three 4,700uF caps per rail.
The rails may sag under load but, by then, the load is loud enough to force some people to cover their ears.
Abs
Ah yes, thanks Abs - I'm sure you're correct; the more complex ideas being discussed in this thread are no doubt very desirable in the right situations, but are probably well into the realms of diminishing returns, or in fact potentially even counterproductive if I (as a noob) tried to impliment them with insufficient understanding of the subtleties. I've also this morning received notice of redundancy, so cost has suddenly become considerably more important than it was yesterday. Thankfully, if I avoid getting carried away, I've got all the really costly components for the build so hopefully i can still carry on with it.
Cheers
kev
Cheers
kev
Twest, thank you once again for the detailed information. Part of my problem is that I read things on the forum and in books but sometimes lack the wider knowledge to put them in context or apply things appropriately, and often where there is disagreement I'm unable to weigh it from an informed position myself. I wasn't aware of the CRC load regulation making it perhaps more suited to class A than B when the load bandwidth goes higher than the cutoff. All in all, and including Abs point and my sudden need to restrain spending, I think a single bank will do the job for me then. Excellent, a final decision I think.
Another example of my noob status is perhaps post #26 - I thought that I'd understood it but i didn't really understand the impact of it in practice. I've already bought some PCBs for project 101 (and the output devices) so i'll probably build that amp as Rod intended initially, which will give me a baseline. If my listening or measuring suggests room for improvement, or even just as an experiment for myself, I'll look then at options for regulation of the low power sections and/or fiddling with the smoothing capacitors.
Cheers
kev
Another example of my noob status is perhaps post #26 - I thought that I'd understood it but i didn't really understand the impact of it in practice. I've already bought some PCBs for project 101 (and the output devices) so i'll probably build that amp as Rod intended initially, which will give me a baseline. If my listening or measuring suggests room for improvement, or even just as an experiment for myself, I'll look then at options for regulation of the low power sections and/or fiddling with the smoothing capacitors.
Cheers
kev
Ouch, sorry to hear of your notice. Understanding at a high level is not the same as understanding implementation tradeoffs; the latter takes more time and ramp up.
I'd suggest planning a balanced input LME49811 build or similar once you're in a better position for it and ABXing it against project 101. Unless one's got a good live sound reference it's often difficult to recognize playback artifacts until listening to a system which manifests them to a different extent, so on the higher performance end of things it can be hard to know what you're missing until hearing it elsewhere. Some folks consider ignorance bliss in this regard, some prefer less accurate playback, some prefer the pro audio sound in = sound out criteria. Mileage also varies with the amp-speaker interaction around cone breakup and voicing but user preferences correlate well with good measured performance once other factors are controlled for. In particular, measuring SPL as a function of amplifier isn't a bad idea. Not many data points available but the ones I'm aware of show 2dB shifts in the highs on similar amp changes. This is usually subjectively perceptible so it can be desirable to control for it.
I'd suggest planning a balanced input LME49811 build or similar once you're in a better position for it and ABXing it against project 101. Unless one's got a good live sound reference it's often difficult to recognize playback artifacts until listening to a system which manifests them to a different extent, so on the higher performance end of things it can be hard to know what you're missing until hearing it elsewhere. Some folks consider ignorance bliss in this regard, some prefer less accurate playback, some prefer the pro audio sound in = sound out criteria. Mileage also varies with the amp-speaker interaction around cone breakup and voicing but user preferences correlate well with good measured performance once other factors are controlled for. In particular, measuring SPL as a function of amplifier isn't a bad idea. Not many data points available but the ones I'm aware of show 2dB shifts in the highs on similar amp changes. This is usually subjectively perceptible so it can be desirable to control for it.
Thanks, its rather worrying but I'm trying to persuade myself that a change would be no bad thing... I looked over the LME49811 data sheet and it seems to have a lot of potential so thanks for bringing it up in this thread, I hadn't realised such things existed. Of course, P101 is already designed, tested and the PCB laid out for me, which is all very reassuring for my first real amp project, but implementing the LME49811 doesn't look too challenging and I suspect it could be a good next step into designing my own PCB, which is quite appealing.
Cheers
Kev
Cheers
Kev
Sorry to hear it. I have friends in the UK who don't speak well of the economy; in fact, they say it is pretty grim. Good luck with finding another job.
On the matter of the LME amps, I did ask Rod about them and he said, in effect, he couldn't see that they offered anything more over his current projects. They may well be a good learning exercise but I would only ask whether the effort is worth it?
Abs
Thanks Abs! The economy here is supposedly picking up now, but personally I've not yet seen much real-world impact so its all very slow. I'm sure I'll manage one way or another, but for the moment it has definately changed how I need to pitch the remaining purchase decisions for this project.
Its interesting to hear what Rod feels about the LME devices. I gather from his description of the P101 amp that he slighly surprised himself with how well it worked out, so I can understand that he'd feel the LMEs wouldn't gain a lot. I'm committed to his amps anyway now and I'm still very happy to have chosen them - not only for the design & PCBs themselves but also for the wealth of information and support that his site provides.
That said, the LMEs seem to offer a useful stepping stone to a home-grown amp should I want to try my own designs in the future, so i can see myself testing them at some point after this project is done. I know theres quite a lot about amp design that I'll not have to deal with by using Rods designs/PCBs/instructions; thats what i wanted in this first/important/costly project, but I'd also like to try learning by my own mistakes on something less critical. Perhaps I'll go on to make some active PC speakers or something, if circumstances permit.
Cheers
kev
Its interesting to hear what Rod feels about the LME devices. I gather from his description of the P101 amp that he slighly surprised himself with how well it worked out, so I can understand that he'd feel the LMEs wouldn't gain a lot. I'm committed to his amps anyway now and I'm still very happy to have chosen them - not only for the design & PCBs themselves but also for the wealth of information and support that his site provides.
That said, the LMEs seem to offer a useful stepping stone to a home-grown amp should I want to try my own designs in the future, so i can see myself testing them at some point after this project is done. I know theres quite a lot about amp design that I'll not have to deal with by using Rods designs/PCBs/instructions; thats what i wanted in this first/important/costly project, but I'd also like to try learning by my own mistakes on something less critical. Perhaps I'll go on to make some active PC speakers or something, if circumstances permit.
Cheers
kev
Can I refer back to post60.
What is the plot showing?
At the left edge where it states music power spectral density exponent = -1
it shows the <300Hz band @ -1.5dB, the > 2k75Hz band at ~-9.8dB and the remaining midband @ ~-7.3dB, relative to total power.
What does that left edge of the plot tell us? what does the -1 mean?
What would change, if the three drivers were all of the same sensitivity instead of the spread of 7dB?
What is the plot showing?
At the left edge where it states music power spectral density exponent = -1
it shows the <300Hz band @ -1.5dB, the > 2k75Hz band at ~-9.8dB and the remaining midband @ ~-7.3dB, relative to total power.
What does that left edge of the plot tell us? what does the -1 mean?
What would change, if the three drivers were all of the same sensitivity instead of the spread of 7dB?
Interesting but not necessarily meaningful without sufficient context to reason about it. The end of "project" 56 says pretty much what I said in post 69; subjective amplifier preferences are (surprise!) subjective. If one's designing for sound in = sound out it's pretty much impossible to beat LME49xxx performance with a discrete implementation. If one's trying to work around, say, cone breakup in a woofer having a tube amp can yield a subjectively preferable result compared to one with a lower output impedance. So can selecting a different driver. Part of the fractious nature of audio stems from a tendency to judge components in isolation rather than considering their interactions.
The relative woofer power will decrease by about 5dB and the tweeter power increase by about 2dB. I'll have to run the plot later but basically what happens is the curves shift to the left. The total power required by a driver is the integral of its output SPL and the PSD applied to it after the crossover. In the case of a reasonably flat mid or tweeter the sensitivity is a good approximation of the SPL over the corresponding crossover passband. For woofers using the nominal sensitivity will underpredict the power required by a couple three dB due to bass rolloff.
1/f = f^-1. 1/sqrt(f) = f^-0.5. More in the second paragraph of post 60.
Yeah, I suppose even an objective opinion can vary hugely depending on the specific circumstances and context, and once you get into subjective opinion then its all very personal. I don't reject subjective opinion, but having worked in/around human sciences for years I'm aware of its limitations and that one needs to employ good methodologies if it is to mean much when extrapolated to a wider audience and/or into other contexts. Similarly i'm sure there are objective measures that are very clear cut but could have considerably less detectable effect on human ears than one may imagine. A minefield, that perhaps ultimately only a given individual can judge for themselves in their own situation!
Cheers
kev
What is the plot showing?
What does the plot tell us?
What does the -1 mean?
Andrew, can you reformulate what it is you're not understanding to ask more in depth questions? I'm struggling, for example, to find a different way of saying -1 on the x axis indicates a 1/f PSD. You're aware that in the small signal model ripple amplitude is the integral of the power consumed by the load over the mains cycle, right?
I suspect for the most part the necessary objective measures exist. However, they're often not in widespread use (Rnonlin, for example) and the subjective correlations with them aren't well understood (the damping factor continuum of a few posts ago as another example). As a result, determining design requirements is difficult as it's hard to predict the point at which increased performance yields no further subjective improvement whilst ensuring appropriate controls for other factors. There's abundant anecdotal evidence listeners place different levels importance on different aspects of sound quality. But there's not, to my knowledge, any sort of n factor model of listener perception the way there are n factor personality models like Big Five NEO PI-R or HEXACO.
The usual way of dealing with this is simply to pursue both one's objective and subjective education by pushing one's design abilities and striving to understand the variables in what one's listening to. Takes a while but the ear's incredibly discerning so it's all rather fascinating.
And that, in a nutshell, is the subjective versus objective "debate" in audio. The medical field is better at quantifying subjective experience than the audio crowd is. Partly due to different ethical requirements but, I think, mostly because the market's substantially larger in monetary terms. As a result, there's a large apparatus of degree programs, residencies, controlled trials, and such to draw from. This doesn't really exist in audio, though the AES tries to some extent. It's also the nature of DIY nearly all experiments lack proper controls due to limited knowledge, money, time, or combination thereof.
I suspect for the most part the necessary objective measures exist. However, they're often not in widespread use (Rnonlin, for example) and the subjective correlations with them aren't well understood (the damping factor continuum of a few posts ago as another example). As a result, determining design requirements is difficult as it's hard to predict the point at which increased performance yields no further subjective improvement whilst ensuring appropriate controls for other factors. There's abundant anecdotal evidence listeners place different levels importance on different aspects of sound quality. But there's not, to my knowledge, any sort of n factor model of listener perception the way there are n factor personality models like Big Five NEO PI-R or HEXACO.
The usual way of dealing with this is simply to pursue both one's objective and subjective education by pushing one's design abilities and striving to understand the variables in what one's listening to. Takes a while but the ear's incredibly discerning so it's all rather fascinating.
Thanks for the link to that student thesis, most interesting I'll have to give it a more detailed read.
Yeah, its very complex and interesting. Another part of the problem with validating objective vs subjectve measurements is that very often the subjective tests will need to be carried out in somewhat artificial conditions (necessary in order to control compounding variables), so that means they too then need validating for relevancy in the real-world.
Cheers
kev
Yeah, its very complex and interesting. Another part of the problem with validating objective vs subjectve measurements is that very often the subjective tests will need to be carried out in somewhat artificial conditions (necessary in order to control compounding variables), so that means they too then need validating for relevancy in the real-world.
Cheers
kev
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.