Interesting the older tpa chip datasheet thd graphs indeed mention even higher value inductors, would they do that if thd deteriorates? It isn't just one Ti application report, they share thd drops with higher inductor values in all those datasheets too, weird.
4th order attenuates switch frequency more, but has maybe even worse amplification result for higher frequencies. Hypex has some papers about that, 400khz isn't very important in any way.
4th order attenuates switch frequency more, but has maybe even worse amplification result for higher frequencies. Hypex has some papers about that, 400khz isn't very important in any way.
You should use a fully differential output filter (2 caps with grounded mid-point) with BD modulation as used as standard in the TPA3116.
The SLOA031 like all other application reports are good resources as long as you take the conditions for them into account. The SLOA031 for example was written in 1999. A lot of things have happened since. Not to theory of design but what is actually achievable in practical designs as components have evolved a lot since then.
Don't forget that back when they were written most people did not consider class D amplifiers to be viable audiophile amplifiers. They were mostly seen as cost/space/energy saving devices that produced a fairly but just barely acceptable result.
Oh boy, have things changed.
The SLOA031 says "The main goal of the output filter is attenuation of the high frequency switching component of the class-D amplifier while preserving the signals in the audio band."
And then goes on to mention that a two pole (2nd order) filter has -40dB attenuation at Fs per decade. Basically cementing the 1:10 golden rule.
You could argue that when you go up in Fs you'd just improve this attenuation and keep the same output filter Fc but there's no need. Modern modulation schemes have built-in EMI suppression that basically eliminates the need for steeper filters and/or increased attenuation. So the golden rule of 1:10 Fc to Fs remains unchanged.
What we instead can do now is move the Fs further away from the audio band which results in vastly increased performance over these earlier designs.
Last edited:
That document seems only focused on AD modulation, where there isn't much common mode frequency content and putting the main output filter capacitor between the two BTL legs is OK. With the BD/1SPW modulation used on the TPA3116/8, you'll potentially create a big EMI mess if you do that, you really need the "big uF" from each BTL leg to ground.
They're making some other questionable recommendations also, eg. using X7R output filter caps. Which is fine, however you have to be aware of the capacitance droop versus frequency that X dielectric caps have, and that it can affect THD performance as well as EMI/carrier rejection (lowering C effectively raises your filter cutoff frequency)
Other than that, don't see anything fundamentally wrong with what they're doing.
And then goes on to mention that a two pole (2nd order) filter has -40dB attenuation at Fs per decade. Basically cementing the 1:10 golden rule.
You could argue that when you go up in Fs you'd just improve this attenuation and keep the same output filter Fc but there's no need. Modern modulation schemes have built-in EMI suppression that basically eliminates the need for steeper filters and/or increased attenuation. So the golden rule of 1:10 Fc to Fs remains unchanged.
What we instead can do now is move the Fs further away from the audio band which results in vastly increased performance over these earlier designs.[/QUOTE]
Does that mean the equations are wrong? As far as I can see they still work to the 1:10 rule, and even using 400khz instead of the 250khz mentioned in slos119b and others they output much the same values noted in slos119b.
I will just have to figure it out when the components arrive.
You could argue that when you go up in Fs you'd just improve this attenuation and keep the same output filter Fc but there's no need. Modern modulation schemes have built-in EMI suppression that basically eliminates the need for steeper filters and/or increased attenuation. So the golden rule of 1:10 Fc to Fs remains unchanged.
What we instead can do now is move the Fs further away from the audio band which results in vastly increased performance over these earlier designs.[/QUOTE]
Does that mean the equations are wrong? As far as I can see they still work to the 1:10 rule, and even using 400khz instead of the 250khz mentioned in slos119b and others they output much the same values noted in slos119b.
I will just have to figure it out when the components arrive.
What is interesting to note that after the Fc is moved an octave out of the audio band there is practically no difference in attenuation with even higher Fc. There is still a good difference in phase degree. And this in my opinion is the key to increased performance. When we know that a decade is the golden rule ratio why are we settling for just an octave between the all important audio band and Fc? The answer is mainly that there aren't that many low µH, high A, low DC ohm inductors intended for the frequency range required out there yet as demand simply hasn't been there. They are slowly emerging though and with the very latest generation of TI class D chips (the new digital only input) with true ternary output modulation they will become much more popular as they allow much higher switching frequencies without severe efficiency penalties.
I think that's my pic (at least I recognize my board holder) from AFTER I had removed the factory caps.
BK
Hypex is, like the first proposed class D amplifier patent back in 1958 it is more or less a direct copy of, a post filter inverting feedback self-oscillating amplifier so that should not come as a surprise to anyone that it is so for Hypex amps. It's also completely irrelevant to this discussion.
Last edited:
Hypex makes really good and massively expensive amps. They operate a very old and tried method of implementing a class D amplifier, and as such it's a good product. The working principle is just fundamentally different than what is discussed in this thread and therefore entirely irrelevant to the discussion.
As you can see, I'm not baited by your pathetic attempts at insulting me. But it would be nice if you come out of that corner you painted yourself into, keep an eye on the ball and engage in proper conversation instead of hurling insults with every post.
As you can see, I'm not baited by your pathetic attempts at insulting me. But it would be nice if you come out of that corner you painted yourself into, keep an eye on the ball and engage in proper conversation instead of hurling insults with every post.
Last edited:
Just download a copy of LTspice and do AC sweeps of whatever filter you come up with, eg for 8 ohms:
OK. Managed to download the software, and then emulated your example to figure out how to use it.it
The graph shows my values hit the fc at 40k, but now I see the peaking clearly. Funnily enough, when I tweaked the cap value to eliminate the peak, the values matched some scribbled notes I made, so I am going to try and piece together the scribbled equation and figure out where I got it.
There is no intended insult.
A tpa3106 isn't much different than a tpa3116, that is what TI is telling designers by pointing at appnotes. The 150khz default increase might slightly better thd performance TIampboards with lower value inductors, thd remarks were starting point. Thd might indeed not be very relevant, thd into a resistor might not be relevant. But TIampboard thd distortion into those resistors is documented.
Frustration at Philips started Hypex, frustration at B&O (Icepower) started some Danish companies. That is a good thing. With most of the modules from any of these companies the speaker has only limited influence on amplifiermodule behavior. Back to TI ampboards and it's feedback design makes things different. Flatening the speakerload or chosing a flat speakerload is best option imo, then you could select LC filterparts.
I intend to hook up my drivers direct, have mocked up a pllxo as lpf to test on 6ohm full range woofer and straight out of box the yj amp performs well enough for the plan. The baffles arent complete yet but i might explore the option of leaving out emi section as amps will be close enough. That's easy to try out as its last mod in chain at output end.
Is the peaking an issue even if its an octave beyond audio range?
Leaving the lc filter aside, when it comes to gain and clipping can somebody be kind enough to explain in laymans terms how this is calculated. I dont quite understand the relationship between the input signal level, the 3volt bias, the gain, and then what the clipping level is. I can't figure if its the voltage at plimmit, virtual rail, or what? Dumb question for someone modding I know.
Is the peaking an issue even if its an octave beyond audio range?
Leaving the lc filter aside, when it comes to gain and clipping can somebody be kind enough to explain in laymans terms how this is calculated. I dont quite understand the relationship between the input signal level, the 3volt bias, the gain, and then what the clipping level is. I can't figure if its the voltage at plimmit, virtual rail, or what? Dumb question for someone modding I know.
Last edited:
20dB gain = 10, or 10 volts on the output per volt on the input. 10^(gain setting in dB / 20) for other settings.
If PLIMIT=GVDD as on most amps, forget about PLIMIT/virtual rail, the output on the amp will clip first.
The maximum output voltage swing of the amp is +-PVCC, so the input voltage swing of the amp is going to be +-(PVCC/10) if you're driving it to the rails with the 20dB gain setting. So for 24V, +-2.4V or +-4.8Vpp or 1.7Vrms sine.
In practice it'll be slightly lower than this - the amp can't drive completely to its rails, but maybe half a volt to them practically. But it's a good starting point.
If PLIMIT=GVDD as on most amps, forget about PLIMIT/virtual rail, the output on the amp will clip first.
The maximum output voltage swing of the amp is +-PVCC, so the input voltage swing of the amp is going to be +-(PVCC/10) if you're driving it to the rails with the 20dB gain setting. So for 24V, +-2.4V or +-4.8Vpp or 1.7Vrms sine.
In practice it'll be slightly lower than this - the amp can't drive completely to its rails, but maybe half a volt to them practically. But it's a good starting point.
If I chose a random input of 1.4Vpp with 20V rail to see if I understand. I get the 26db gain setting. Is that right?
I'm going crazy about this. This is my second attempt to make pbtl mode tpa3116. I used YJ-blue black boards at first. Tried one ps to feed both boards resulted scratchy distortion, even with one ps for one board. So I ordered audiobah clones from YJ. Modded for bootstrap (grounded at the ps caps ground pin), replaced ps caps with silmic and followed instructions for pbtl.
Result ; silence!!!!
I'm powering boards with seperated ps.
Check image file please;
Is it wrong implementetion? As I read, solder pad close to signal input must be grounded alltogether. Middle ones for bootstrap mode, near output pads must be joined for pbtl too..
What am I doing wrong?!
Result ; silence!!!!
I'm powering boards with seperated ps.
Check image file please;
Is it wrong implementetion? As I read, solder pad close to signal input must be grounded alltogether. Middle ones for bootstrap mode, near output pads must be joined for pbtl too..
What am I doing wrong?!
I'm going crazy about this. This is my second attempt to make pbtl mode tpa3116. I used YJ-blue black boards at first. Tried one ps to feed both boards resulted scratchy distortion, even with one ps for one board. So I ordered audiobah clones from YJ. Modded for bootstrap (grounded at the ps caps ground pin), replaced ps caps with silmic and followed instructions for pbtl.
Result ; silence!!!!
I'm powering boards with seperated ps.
Check image file please;
An externally hosted image should be here but it was not working when we last tested it.
Is it wrong implementetion? As I read, solder pad close to signal input must be grounded alltogether. Middle ones for bootstrap mode, near output pads must be joined for pbtl too..
What am I doing wrong?!
Is this board came in in PBTL mode already?
The "original" Audiobah board that I had were in stereo mode. There were 5 solder pads (I think) that you have to link them to turn it into PBTL mode. Looking at your board, I did not see that.
Regards,