On the question of the LC filter inductor coils, can I ask as to why there seems little mention of changing the values of the filter capacitor.
I agree that increasing the size of the coils from the low values fitted by the suppliers is a good idea, but without altering the capacitor the change of coil shifts the frequency cut off, and depending on the ohms of speakers used it can also be detrimental to the performance.
I couldn't imagine anyone would switch inductor coil values in a crossover without tweaking the capacitor values also.
From various resources I calculated that for an 8ohm load with 400khz switching freq the coil is optimised at 22uh with a 0.68uF capacitor. The same calculations for the 6ohm load gives 16uh coil and 1uF capacitor (both of these rounded to the nearest values. The 6ohm is actually 16.8uh with 0.96uf cap). The calculation for 4ohm load changes again.
The above give a frequency cut off just above 41khz and just below, respectively. When I cross referenced the results with the Ti info on class d filter design the 8ohm results gave lowest THD of 0.05 in their tables. The 6ohm results gave THD of 0.09.
The equation I used bases the switching freq at 400khz, and sets -40db per decade. Using the 41khz as the freq cutoff. I read that a problem with the freq cutoff being too low is that although the low frequency knee is fine, it can introduce the high frequency knee to audible levels. Another resource on class d design suggested the freq cut off optimum between 30k - 60kohm. Whatever the chosen freq cutoff the capacitor value needs consideration when raising inductor coil values.
I realise I've rambled a bit. Just slinging in my two penneth.
I agree that increasing the size of the coils from the low values fitted by the suppliers is a good idea, but without altering the capacitor the change of coil shifts the frequency cut off, and depending on the ohms of speakers used it can also be detrimental to the performance.
I couldn't imagine anyone would switch inductor coil values in a crossover without tweaking the capacitor values also.
From various resources I calculated that for an 8ohm load with 400khz switching freq the coil is optimised at 22uh with a 0.68uF capacitor. The same calculations for the 6ohm load gives 16uh coil and 1uF capacitor (both of these rounded to the nearest values. The 6ohm is actually 16.8uh with 0.96uf cap). The calculation for 4ohm load changes again.
The above give a frequency cut off just above 41khz and just below, respectively. When I cross referenced the results with the Ti info on class d filter design the 8ohm results gave lowest THD of 0.05 in their tables. The 6ohm results gave THD of 0.09.
The equation I used bases the switching freq at 400khz, and sets -40db per decade. Using the 41khz as the freq cutoff. I read that a problem with the freq cutoff being too low is that although the low frequency knee is fine, it can introduce the high frequency knee to audible levels. Another resource on class d design suggested the freq cut off optimum between 30k - 60kohm. Whatever the chosen freq cutoff the capacitor value needs consideration when raising inductor coil values.
I realise I've rambled a bit. Just slinging in my two penneth.
As a general statement this is incorrect. The ratio between switching frequency and filter frequency should follow the golden rule and be 1:10. You have a little room to maneuver but generally speaking it should be within 1:8 and 1:12.
If there is any improvement by changing to a higher value inductor for a given load impedance then it is because the inductors fitted does not hold their nominal inductance in actual use due to saturation or just poor quality.
Changing the switching frequency will allow you to use smaller value inductors as you can and should use a higher filter frequency to maintain the golden ratio.
All things being equal, a smaller value inductor of the same physical size will have higher Ampere saturation and lower DC resistance which in turn translates to better performance.
Last edited:
and higher thd
No. Lower THD.
Not that it really matters too much what the THD of the amp is. If it's 0.2% or 0.02% doesn't really make that much a difference compared to the effects on THD of the speakers and especially the room acoustics.
Seems you still don't get what the Application Report you desperately cling to actually says.
What is shown in that is that it is better from THD perspective to maintain the golden ratio between switching frequency and filter frequency, and then allow peaking in the filter to linearize the output.
So yes, it is quite clever actually in that particular application where a switching frequency is as low as 250khz. Forcing the filter frequency to be no higher than the about 28khz it is with a 33µH inductor on that particular load impedance if the golden ratio has to be maintained. Naturally a low 28khz filter frequency two pole filter will have effects in the audible range but he eliminates that by adding peaking. Adding peaking will naturally add distortion as expressed by THD but what he shows is that the reduced THD by maintaining the golden ratio between switching frequency and filter frequency is of larger significance than the increased THD by adding peaking.
Here we're talking about an amplifier chip that has 400khz minimum switching frequency, and arguably performs best at 500khz-600khz so the application is completely different, and therefore the Application Report you refer to is (mostly) invalid for this chip.
What is shown in that is that it is better from THD perspective to maintain the golden ratio between switching frequency and filter frequency, and then allow peaking in the filter to linearize the output.
So yes, it is quite clever actually in that particular application where a switching frequency is as low as 250khz. Forcing the filter frequency to be no higher than the about 28khz it is with a 33µH inductor on that particular load impedance if the golden ratio has to be maintained. Naturally a low 28khz filter frequency two pole filter will have effects in the audible range but he eliminates that by adding peaking. Adding peaking will naturally add distortion as expressed by THD but what he shows is that the reduced THD by maintaining the golden ratio between switching frequency and filter frequency is of larger significance than the increased THD by adding peaking.
Here we're talking about an amplifier chip that has 400khz minimum switching frequency, and arguably performs best at 500khz-600khz so the application is completely different, and therefore the Application Report you refer to is (mostly) invalid for this chip.
Last edited:
http://www.ti.com/lit/gpn/tpa3116d2
Figure 2 and 4
Figure 3 and 5
And for the record since you have an obvious problem understanding this. I'm not stating TI is wrong. In fact the opposite. I'm stating that you do not understand why TI is right and therefore have drawn the wrong conclusion that I must somehow be in disagreement with TI when in fact I am not.
Now. Can we please get back on topic.
Last edited:
I confess to not understanding anywhere near as much as others regarding the subject. My experience is based on the 10uh coils of the yj blue/black boards I received, and the equations in Ti sloa031 and 119B. Also considering Ti sloa 201 and published snubber paper by Jim hagermann.
If I remember correctly I worked out the 10uh with 1uf filter values and it put the fc well below 18khz. I only checked the 22uh/680nf values on Ti sheets after using equations to calculate them using my load and 400khz. I must correct my previous THD value, it is 0.075 for the 22/0.68. The 0.05 is achievable with 33uh/1uf but with a 28khz fc. This is for 8ohm load.
The 10uh/1uf which came with the boards are optimum for 4ohm loads, giving a thd of 0.082 at 48khz fc. This happens to be the best thd of the 4ohm chart.
My main point was meant to be that many suggestions of changing coils to different values have been made, but as far as I have seen the suggestion never mentions the capacitor. If these suggestions are for the 3116 with a 400khz switching frequency then as you say the best fc is 40khz (1:10). But changing coils without changing capacitors does not adhere to this fc. By also changing the capacitor it is possible to match the fc, as close as possible anyway.
Anyhoo... I bow to all those modding these boards. In standard form mine sound good, with parts on the way I hope them to sound fantastic. Time, new batteries in the calculator, and a start hand will tell. ;-)
If I remember correctly I worked out the 10uh with 1uf filter values and it put the fc well below 18khz. I only checked the 22uh/680nf values on Ti sheets after using equations to calculate them using my load and 400khz. I must correct my previous THD value, it is 0.075 for the 22/0.68. The 0.05 is achievable with 33uh/1uf but with a 28khz fc. This is for 8ohm load.
The 10uh/1uf which came with the boards are optimum for 4ohm loads, giving a thd of 0.082 at 48khz fc. This happens to be the best thd of the 4ohm chart.
My main point was meant to be that many suggestions of changing coils to different values have been made, but as far as I have seen the suggestion never mentions the capacitor. If these suggestions are for the 3116 with a 400khz switching frequency then as you say the best fc is 40khz (1:10). But changing coils without changing capacitors does not adhere to this fc. By also changing the capacitor it is possible to match the fc, as close as possible anyway.
Anyhoo... I bow to all those modding these boards. In standard form mine sound good, with parts on the way I hope them to sound fantastic. Time, new batteries in the calculator, and a start hand will tell. ;-)
I concur. The capacitor in the output filter should naturally be optimized for the filter frequency as well. This is often not mentioned when people are talking about different output inductor values but it is implied.
In all this it is also important to note that everything depends on load impedance as well and since that usually varies quite a lot it is really doesn't make sense for us to talk about any one standard filter. Instead we should see them as optimum starting points for modification in order to suit the specific speakers the amp is intended to drive.
Naturally you can brute force it by using impedance correction but that's really not a good solution at all. It is far better to optimize the amp for the speaker instead of trying to optimize the speaker for the amp.
It it also interesting to note that we're many times talking about differences in THD that are like hearing the difference between the sneeze or the cough of a flea on a donkeys behind at 200 feet.
In all this it is also important to note that everything depends on load impedance as well and since that usually varies quite a lot it is really doesn't make sense for us to talk about any one standard filter. Instead we should see them as optimum starting points for modification in order to suit the specific speakers the amp is intended to drive.
Naturally you can brute force it by using impedance correction but that's really not a good solution at all. It is far better to optimize the amp for the speaker instead of trying to optimize the speaker for the amp.
It it also interesting to note that we're many times talking about differences in THD that are like hearing the difference between the sneeze or the cough of a flea on a donkeys behind at 200 feet.
Loads of silly people
(link doesn't work btw) 3116 1khz thd is higher than 3110/3123, even with 66uH or 47uH for the last two. Without this added distortion by the huge inductors the old TI chips would totally degrade all new ones lol. I only mention the thd as it highlights to lower value matches the proximity to the 40khz fc.
I am actually quite pleased that I received a response from yourself Saturnus. Have you any advise on if a pllxo is possible before the amp for bi-amping? I have considered a bandpass 4pole shunt incorporating the input DC coupling as the final cap for a full range driver. And using a 3 pole hpf for tweeter incorporating the input coupling as final cap. Would this be likely possible?
I am actually quite pleased that I received a response from yourself Saturnus. Have you any advise on if a pllxo is possible before the amp for bi-amping? I have considered a bandpass 4pole shunt incorporating the input DC coupling as the final cap for a full range driver. And using a 3 pole hpf for tweeter incorporating the input coupling as final cap. Would this be likely possible?
Anything above 2 pole pllxo is not advisable in my opinion. If you need higher order you can always calc the correct frequency values for the different poles but apply the last poles on the output side of the amp.
For example a 4th order passive filter can be realized by having the first 2 passive poles on the input side and the last 2 passive poles on the output side.
For example a 4th order passive filter can be realized by having the first 2 passive poles on the input side and the last 2 passive poles on the output side.
Would you stop bringing up slos119b?and higher thd
Different inductors are used, and no inductance vs current plot is available for any of them, plus the inductors are run well past their current ratings in the test setup. This can very well cause a great deal of THD by itself.
Ceramic capacitors are also used - interestingly, capacitance droop with voltage can correct the inductance droop with current, causing the damping factor to remain more consistent with output swing. So an inductor that suffers more L droop may actually provide a better THD measurement if it matches the C droop.
Lastly it's a TPA3106 used there, which is a different chip running at a lower frequency. Its open loop gain and compensation are likely set lower than the TPA3116/8 chip, reducing its ability to correct THD via feedback, and changing its effective "sweet spot"
Overall, you can't look at that paper and say "higher L = lower THD" in every other amplifier design. There's too many other things at play in that experiment to draw such a conclusion. Throw that paper out.
S
Mmmmm. I am new to using the term "poles" so I dont know if what i am suggesting is a 2 or 4 pole.
Basically a bandpass shunt made of inductor and cap in series with same in parallel. I was thinking the coupling cap could be the series cap. And if possible using a 3 component hpf on other channel for tweeter, using coupling cap. If possible it allows me to keep same value coupling cap on both channels.
Just to note, I am as cautious about the graphs in the tpa3116 datasheet you mentioned as you sound about my doc with the the values. The TPA datasheet makes no mention of changing the lc filter in obtaining the 4ohm and 8ohm results, so one probably isnt optimized for load. Also does mention any component adjustment for change 12V to 24V either. I know to be cautious with the datasheets as throughout the class d filter doc it only mentions 250khz switching. That confused me for a bit!
Mmmmm. I am new to using the term "poles" so I dont know if what i am suggesting is a 2 or 4 pole.
Basically a bandpass shunt made of inductor and cap in series with same in parallel. I was thinking the coupling cap could be the series cap. And if possible using a 3 component hpf on other channel for tweeter, using coupling cap. If possible it allows me to keep same value coupling cap on both channels.
Just to note, I am as cautious about the graphs in the tpa3116 datasheet you mentioned as you sound about my doc with the the values. The TPA datasheet makes no mention of changing the lc filter in obtaining the 4ohm and 8ohm results, so one probably isnt optimized for load. Also does mention any component adjustment for change 12V to 24V either. I know to be cautious with the datasheets as throughout the class d filter doc it only mentions 250khz switching. That confused me for a bit!
I'm assuming it's the "Outputfilter:L=10μH,C=680nF" mentioned in the "recommended operating conditions", as indicated in the "typical application" diagram, and as used on the TPA3116 evaluation board, so it's really not optimized for neither a 4 ohm nor a 8 ohm load. However, I cannot be absolutely certain that they have not used different output filter configurations optimized for the different loads in the data sheet but I'd be very surprised if they have.
What's your judgement of SLOA031? That's the doc I actually used equations from to calculate the lc filter values I mentioned. It just so happened that they matched the sloa119 when I cross checked them.
The equation used from sloa031 for lc filter is:
Capacitor= 1÷(2×π×fc×√2×1/2 load in ohms)
Inductor= √2×1/2 load in ohms ÷(2×π×fc)
To check the fc of the filter use equation in there:
Fc= 1÷(2×π×√(2×inductor×(capacitor÷2))). The reason its capacitor ÷2 is that the equation
In doc is for a btl with a single cap instead of 2. Basically requiring you to only put the value of 1 cap, not both, into the equation.
Last edited: