I have just finished my first DIY dual mono amp, is is in bridged lm3886 config. The case i made myself with 5mm thick alluminium and heatsinks as walls. The design i followed is from the national datasheet and shown below. I have changed Rf to make the amplifier gain around 30. I am getting great sound out of it, zero hum or noise that i can hear. The only complaint is that it does not seem to have very much bass. I may possibly increase the gain furthur as some sources i have to turn volume control right up to get loud volume. Previously i have just changed Rf to vary the gain, should i change Rin as well to keep things ballanced?I would like any suggestions on ways to improve my amp.
This is the power supply design i have used, I am using +-35 volt supplies, from dual 25-0-25 toroids. 10,000uF of supply capacitance per rail for each channel. do you think any performance benifets will be achieved if i add another 10,000uF to each rail. I also have 1uF accross the supply rails at the chips, should i have 0.1uf as well?
These are photos of the finished amp:
An externally hosted image should be here but it was not working when we last tested it.
This is the power supply design i have used, I am using +-35 volt supplies, from dual 25-0-25 toroids. 10,000uF of supply capacitance per rail for each channel. do you think any performance benifets will be achieved if i add another 10,000uF to each rail. I also have 1uF accross the supply rails at the chips, should i have 0.1uf as well?
An externally hosted image should be here but it was not working when we last tested it.
These are photos of the finished amp:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Hi,
you have different gain in each half of the bridge. 14.1 Vs 14.5
If you select each resistor @~+-0.5% from nominal you can get the gains very close.
The turn over frequency (F-3db) of the non-inverting is about 3.4Hz, but the inverting F-3db is 7.2Hz. Could this be the bass loss problem?
On it's own this is probably quite subtle.
Agh! the non-inverting channel has the NFB F-3db set higher than the input filter (F-3db~=7.2Hz). This should be set about half to one octave LOWER than the input filter.
Each channel has Zin~=4k3 this is unusually low, could the source be struggling to drive this?
But it gets worse, the two bridged amps feed ~100W into 8ohms. A normal power amp of this capability would require +-10mF to +-20mF on the rails of each channel.
But each half of the bridged amp acts as a 50W into 4ohm power amplifier. To produce good bass into that load you need +-20mF to +-40mF per amplifier. But you have two amplifiers each producing 50W into 4ohms working in unison. So your smoothing capacitance for the pair jumps to +-40mF to +-80mF per channel.
You have just +-10mF per channel.
That is one of the major difficulties/costs that is generally ignored when deciding to adopt bridged amplifiers.
This shortfall in smoothing capacitance combined with the highish F-3db could explain the bass loss.
I suggest that instead, you convert each channel to parallel operation. Each amp will try to drive 25W into 16ohms if you retain your 8ohm speakers. Change the turnover frequency to below 2Hz in the non-inverting mode and fit +-20mF to each parallel chipamp channel.
This will make a much better balanced amp capable of delivering as much current as any 8ohm speaker could ever require. It would even make a pretty good stab at running 4 to 8ohm speakers. You gain all this for a loss of 3db in maximum volume.
you have different gain in each half of the bridge. 14.1 Vs 14.5
If you select each resistor @~+-0.5% from nominal you can get the gains very close.
The turn over frequency (F-3db) of the non-inverting is about 3.4Hz, but the inverting F-3db is 7.2Hz. Could this be the bass loss problem?
On it's own this is probably quite subtle.
Agh! the non-inverting channel has the NFB F-3db set higher than the input filter (F-3db~=7.2Hz). This should be set about half to one octave LOWER than the input filter.
Each channel has Zin~=4k3 this is unusually low, could the source be struggling to drive this?
But it gets worse, the two bridged amps feed ~100W into 8ohms. A normal power amp of this capability would require +-10mF to +-20mF on the rails of each channel.
But each half of the bridged amp acts as a 50W into 4ohm power amplifier. To produce good bass into that load you need +-20mF to +-40mF per amplifier. But you have two amplifiers each producing 50W into 4ohms working in unison. So your smoothing capacitance for the pair jumps to +-40mF to +-80mF per channel.
You have just +-10mF per channel.
That is one of the major difficulties/costs that is generally ignored when deciding to adopt bridged amplifiers.
This shortfall in smoothing capacitance combined with the highish F-3db could explain the bass loss.
I suggest that instead, you convert each channel to parallel operation. Each amp will try to drive 25W into 16ohms if you retain your 8ohm speakers. Change the turnover frequency to below 2Hz in the non-inverting mode and fit +-20mF to each parallel chipamp channel.
This will make a much better balanced amp capable of delivering as much current as any 8ohm speaker could ever require. It would even make a pretty good stab at running 4 to 8ohm speakers. You gain all this for a loss of 3db in maximum volume.
Thanks for your comments
I am confused, i know The input filter is calculated from Rin and Cin, and equates to 3.38Hz.
The -3db point for the inverting opamp is calculated from Ri2 and Ci2 and equates to 7.2Hz
But i am a little confused on how you calculate the non-inverting half power point, should it not be the same as the inverting opamp but using Ri1 and Ci1 ??
If i increase Ri1 and Ri2 to a value of around 22K this will increase the input impedance substantially and also drop -3db point to around 1.5Hz.........Right?..........Of course i will have to increase Rf to keep the gain around 30.
Could you please point me in the right directions and tell me what components i need to change to get the -3db cutoff to below 2 or 3Hz.
I have no trouble in adding more PS capacitance as i had thought of doing so anyway. I will start by adding another 10,000 to each rail and see if bass improves.
I am confused, i know The input filter is calculated from Rin and Cin, and equates to 3.38Hz.
The -3db point for the inverting opamp is calculated from Ri2 and Ci2 and equates to 7.2Hz
But i am a little confused on how you calculate the non-inverting half power point, should it not be the same as the inverting opamp but using Ri1 and Ci1 ??
If i increase Ri1 and Ri2 to a value of around 22K this will increase the input impedance substantially and also drop -3db point to around 1.5Hz.........Right?..........Of course i will have to increase Rf to keep the gain around 30.
Could you please point me in the right directions and tell me what components i need to change to get the -3db cutoff to below 2 or 3Hz.
I have no trouble in adding more PS capacitance as i had thought of doing so anyway. I will start by adding another 10,000 to each rail and see if bass improves.
Hi,
the input filters should be the restriction to the bandwidth not the NFB leg.
For the inverting using 4u7F as the DC blocking cap, the input resistor could be changed to 20k to get 90mS (f-3db=1.8Hz) RC time constant. But this needs a very high value of feedback (300k) resistor.
You could instead fit a parallel combination of 2u2F//47uF and now you can use 1k8 to give the same 90mS RC. The feedback now becomes 27k. Zin<1k8 and low noise.
But I see non-inverting as the better option if your source cannot drive 1k8 or 4k7 adequately.
For the non-inverting input, you can use 2u2 and 39k or 43k for the input filter.
the feed back then becomes 39k & 2k7 for that same 14times gain.
The NFB filter should be set at least half an octave below the 1.8Hz high pass at the input. try to get RC about 130mS to 150mS here.
Finally for a 4ohm load (that's what the bridged pair each see) the PSU RC>=1.4 *140mS = ~200mS, that amounts to +-50mF on every amplifier rail. This bridging is getting silly. Go parallel with non-inverting.
At least try it on one channel since you already have all the components and listen to the bass (if any) difference in the two channels. The ease with which the parallel drives your 8ohm speakers may sound noticeable in the midrange as well.
Remember to match the gain to about 30 in both channels if you go for this experiment.
the input filters should be the restriction to the bandwidth not the NFB leg.
For the inverting using 4u7F as the DC blocking cap, the input resistor could be changed to 20k to get 90mS (f-3db=1.8Hz) RC time constant. But this needs a very high value of feedback (300k) resistor.
You could instead fit a parallel combination of 2u2F//47uF and now you can use 1k8 to give the same 90mS RC. The feedback now becomes 27k. Zin<1k8 and low noise.
But I see non-inverting as the better option if your source cannot drive 1k8 or 4k7 adequately.
For the non-inverting input, you can use 2u2 and 39k or 43k for the input filter.
the feed back then becomes 39k & 2k7 for that same 14times gain.
The NFB filter should be set at least half an octave below the 1.8Hz high pass at the input. try to get RC about 130mS to 150mS here.
Finally for a 4ohm load (that's what the bridged pair each see) the PSU RC>=1.4 *140mS = ~200mS, that amounts to +-50mF on every amplifier rail. This bridging is getting silly. Go parallel with non-inverting.
At least try it on one channel since you already have all the components and listen to the bass (if any) difference in the two channels. The ease with which the parallel drives your 8ohm speakers may sound noticeable in the midrange as well.
Remember to match the gain to about 30 in both channels if you go for this experiment.
I second Andrews remarks and suggest you take a look at this application note dealing with details of bridging/paralleling LM3886's:
http://www.national.com/an/AN/AN-1192.pdf
Regards, Klaus
http://www.national.com/an/AN/AN-1192.pdf
Regards, Klaus
Hi - why would i change parallel capacitors and change Ri2 to 1.8k, i would think i should be increasing the Inverting input impedance.
"But I see non-inverting as the better option if your source cannot drive 1k8 or 4k7 adequately."
please explain this, how is it the better option, what is the other option?
"The NFB filter should be set at least half an octave below the 1.8Hz high pass at the input. "
I though the input high pass cut off was 3.4hZ
In regards to reading the application document, i have and that is where this schematic if from. Are you saying that the national schematic, may have significant flaws in the design.
"But I see non-inverting as the better option if your source cannot drive 1k8 or 4k7 adequately."
please explain this, how is it the better option, what is the other option?
"The NFB filter should be set at least half an octave below the 1.8Hz high pass at the input. "
I though the input high pass cut off was 3.4hZ
In regards to reading the application document, i have and that is where this schematic if from. Are you saying that the national schematic, may have significant flaws in the design.
Hi Scott,
The basic circuit you have built assumes a low driving impedance, otherwise the gain of inverting side will vary with source impedance. This will be significant if you have a level control pot in front of the amp (a 10k pot has 2.5k max. impedance). See circuit fig.17, there they have a buffer installed and the resistors scaled down, capacitors scaled up. I would suggest you build that circuit (omitting the paralleled sections).
OTOH, not reaching 100% perfect gain matching is not too much of a real problem, the output will just not be pefectly symmetrical. That would take away a little of max. reachable output voltage/power.
I would opt for the high-pass corner frequencies in the feedback sections to be at 3Hz, for example. Then the input RC for the non-inverting section should be at 0.75Hz (2 octaves down) so that it will not contribute much to the total roll-off. If it is at higher frequencies, the noninverting section will see additional roll-off and phase shift, both sections wouldn't be balanced any more for very low frequencies. Isn't too much of a significant problem, though.
Maybe more important -- and that is a point where the schematic fig.3 looks flawed indeed -- is the matching of the DC impedances at all op-amp inputs, to get low offsets. For DC the caps are open circuit, so the conditions are: Rin=Rf1, Rb=Rf2. A low value of Rb would be correct only if Ci2 were shorted and the source has a low impedance DC-coupled output. Actually it would be 4.3k, not 3.32k then (the 3.32k might be an empirical value for the actual circuit that was choosen to minmize offset). Again, fig.17 looks better in this regard.
Like has been said, the restrictions of the chip with 4 ohm loads would suggest you go parallel instead, like in fig.6. You will be rewarded with sound quality, the small loss in output power is only a very little price to pay for that. And in the non-inverting mode you can easily get a high input impedance (and use smaller coupling caps).
Those datasheet/appnote schematics should always be seen as starting points, not as bullet-proof and ready-to-build matured circuits.
Regards, Klaus
The basic circuit you have built assumes a low driving impedance, otherwise the gain of inverting side will vary with source impedance. This will be significant if you have a level control pot in front of the amp (a 10k pot has 2.5k max. impedance). See circuit fig.17, there they have a buffer installed and the resistors scaled down, capacitors scaled up. I would suggest you build that circuit (omitting the paralleled sections).
OTOH, not reaching 100% perfect gain matching is not too much of a real problem, the output will just not be pefectly symmetrical. That would take away a little of max. reachable output voltage/power.
I would opt for the high-pass corner frequencies in the feedback sections to be at 3Hz, for example. Then the input RC for the non-inverting section should be at 0.75Hz (2 octaves down) so that it will not contribute much to the total roll-off. If it is at higher frequencies, the noninverting section will see additional roll-off and phase shift, both sections wouldn't be balanced any more for very low frequencies. Isn't too much of a significant problem, though.
Maybe more important -- and that is a point where the schematic fig.3 looks flawed indeed -- is the matching of the DC impedances at all op-amp inputs, to get low offsets. For DC the caps are open circuit, so the conditions are: Rin=Rf1, Rb=Rf2. A low value of Rb would be correct only if Ci2 were shorted and the source has a low impedance DC-coupled output. Actually it would be 4.3k, not 3.32k then (the 3.32k might be an empirical value for the actual circuit that was choosen to minmize offset). Again, fig.17 looks better in this regard.
Like has been said, the restrictions of the chip with 4 ohm loads would suggest you go parallel instead, like in fig.6. You will be rewarded with sound quality, the small loss in output power is only a very little price to pay for that. And in the non-inverting mode you can easily get a high input impedance (and use smaller coupling caps).
Those datasheet/appnote schematics should always be seen as starting points, not as bullet-proof and ready-to-build matured circuits.
Regards, Klaus
Hi,
decide how low you want/need the bass response to go.
Set all the input filters to that frequency.
Set all the NFB filters to half an octave below your chosen input filter frequency.
Set the PSU RC to half an octave below the NFB frequency.
Now see what components should be changed to and what extra components are needed to produce your desired roll-off frequency.
Now redo all the above for a parallel arrangement.
which is cheaper and easier to accomplish?
which is likely to play/sound better?
which is 3db down on power output?
the answer to all three should be parallel.
BTW,
I have found that changing the input filter from F-3db=10Hz to F-3db=2Hz makes a significant difference to the way small bookshelf speakers are able to play bass. Changing from 6Hz to 2Hz is also noticeable but sutble rather than significant. Using big speakers have an even greater effect.
I very much prefer setting the input filter to about 80 to 90mS (1.8Hz) and then choosing all the others appropriately.
decide how low you want/need the bass response to go.
Set all the input filters to that frequency.
Set all the NFB filters to half an octave below your chosen input filter frequency.
Set the PSU RC to half an octave below the NFB frequency.
Now see what components should be changed to and what extra components are needed to produce your desired roll-off frequency.
Now redo all the above for a parallel arrangement.
which is cheaper and easier to accomplish?
which is likely to play/sound better?
which is 3db down on power output?
the answer to all three should be parallel.
BTW,
I have found that changing the input filter from F-3db=10Hz to F-3db=2Hz makes a significant difference to the way small bookshelf speakers are able to play bass. Changing from 6Hz to 2Hz is also noticeable but sutble rather than significant. Using big speakers have an even greater effect.
I very much prefer setting the input filter to about 80 to 90mS (1.8Hz) and then choosing all the others appropriately.
Hi,
the input filters are Cin, Rin=1uF & 47K=47mS on the n-inv
and Ci2, Ri2=4u7F & 4k7=22mS
the NFB filter is Ci1, Ri1=4u7F & 4k7=22mS
Assuming that the preamp/source has Rs<=100r (Rs=51r would be better) and can drive 1k0 to +20dbu (~7.7Vrms) requiring output current of 11mApk +capacitance drive, so say one needs +-20mApk.
We can select input filter to 90mS
Cin=2u2F, Rin=40k9 use 39k or 43k.
Ci2=47uF//2u2, Ri2=1k8, Rf2=27k, Rb=27k(+trimmer?)
NFB filter set to 130mS
Rf1=Rin=39k, Ri1=39k/14=2k786 (2k7+82r or 3k//39k), Ci1=46.7uF use 47uF//2u2F
gain approx 30times (+29.5db)
Zin~=1k7
Rs<=100r
You cannot use an unbuffered source with this arrangement.
Each amplifier requires PSU RC=184mS, R=4ohm, C=+-46mF
Notice that choosing 90mS has determined every component connected to the chipamps. They are all inter-related.
Now consider the parallel non-inverting arrangement.
set input filter to 90mS
Cin=2u2F, Rin=39k
Rf1=39k, Ri1=39k/29=1k34 use 1k3, Ci1=100uF//2u2F
PSU RC=188mS, R=16ohms, C=+-11.7mF, use +-12mF on each amplifier.
Zin=19k5
Rs<=1k +20dbu requiring current output of 0.6mApk (ClassA single ended buffer is OK for this)
This arrangement could be made to work with an unbuffered source using a 5k pot as volume control.
Both versions of this circuit needs RF filtering on the inputs.
Consider seriously before omitting it.
I wonder what my fee should be?
The rail fuses after the smoothing caps can be half the peak current from the amps into their nominal load.
Bridged needs T3.1A, parallel needs T2A using 8ohm speakers.
If you use +-1m5F at the chip these fuses may suffer nuisance blowing at start up. Be prepared to increasing them slightly.
There was a lot of back of a fag packet calculating in this reply.
Would anyone care to check my numbers? I don't want someone buying the wrong component values.
Remember, this all started with choosing the high pass filter frequency of the amplifier. These numbers can be scaled to any frequency, 90mS happens to suit (in my view) playing bass down to 20Hz.
the input filters are Cin, Rin=1uF & 47K=47mS on the n-inv
and Ci2, Ri2=4u7F & 4k7=22mS
the NFB filter is Ci1, Ri1=4u7F & 4k7=22mS
Assuming that the preamp/source has Rs<=100r (Rs=51r would be better) and can drive 1k0 to +20dbu (~7.7Vrms) requiring output current of 11mApk +capacitance drive, so say one needs +-20mApk.
We can select input filter to 90mS
Cin=2u2F, Rin=40k9 use 39k or 43k.
Ci2=47uF//2u2, Ri2=1k8, Rf2=27k, Rb=27k(+trimmer?)
NFB filter set to 130mS
Rf1=Rin=39k, Ri1=39k/14=2k786 (2k7+82r or 3k//39k), Ci1=46.7uF use 47uF//2u2F
gain approx 30times (+29.5db)
Zin~=1k7
Rs<=100r
You cannot use an unbuffered source with this arrangement.
Each amplifier requires PSU RC=184mS, R=4ohm, C=+-46mF
Notice that choosing 90mS has determined every component connected to the chipamps. They are all inter-related.
Now consider the parallel non-inverting arrangement.
set input filter to 90mS
Cin=2u2F, Rin=39k
Rf1=39k, Ri1=39k/29=1k34 use 1k3, Ci1=100uF//2u2F
PSU RC=188mS, R=16ohms, C=+-11.7mF, use +-12mF on each amplifier.
Zin=19k5
Rs<=1k +20dbu requiring current output of 0.6mApk (ClassA single ended buffer is OK for this)
This arrangement could be made to work with an unbuffered source using a 5k pot as volume control.
Both versions of this circuit needs RF filtering on the inputs.
Consider seriously before omitting it.
I wonder what my fee should be?
The rail fuses after the smoothing caps can be half the peak current from the amps into their nominal load.
Bridged needs T3.1A, parallel needs T2A using 8ohm speakers.
If you use +-1m5F at the chip these fuses may suffer nuisance blowing at start up. Be prepared to increasing them slightly.
There was a lot of back of a fag packet calculating in this reply.
Would anyone care to check my numbers? I don't want someone buying the wrong component values.
Remember, this all started with choosing the high pass filter frequency of the amplifier. These numbers can be scaled to any frequency, 90mS happens to suit (in my view) playing bass down to 20Hz.
i think the way is no good to building a bridge amp.because we talk about is building a real amp.
input filters and nfb input filters will make things going to complicate,the unstandard caps and resistors is a shopping hell.
i agreed with AndrewT.the parallel building is a good solution of an1192.otherwise you shoudl make a bridge driver to handle it.
zang
input filters and nfb input filters will make things going to complicate,the unstandard caps and resistors is a shopping hell.
i agreed with AndrewT.the parallel building is a good solution of an1192.otherwise you shoudl make a bridge driver to handle it.
zang
Hi,
@Andrew: Like the old saying goes "two engineers will have five opinions", I see some things a little different...
There is only one input filter Cin, Rin which is effective only for the non-inv, and two filters in the feedbacks. The feedback filters should be matched, while the input filter should not add additional roll-off and phase. Exact matching is impossible, though, because the non-inv drops to gain=1 for DC while the inv drops to zero. In the extreme case of very little gain, say 2dB, there is no -3dB roll-off to be reached for the non inv.
There is a big gotcha in the schem, because if the input is left open, the ac noise gain of the inv is 68k/(47k+4.7k)=1.3 which quite certainly causes the inv to oscillate, given the phase margin of the chip. A buffer is necessary, therefore.
After a little calculating and simulating I arrived to the following values:
Rin=47k, Cin=47uF ==> Fc=0.072Hz, This RC is before the buffer, of course.
Ri1=Ri1=4.7k, Ci1=Ci2=47uF ==> Fc=~0.72Hz
Rf1=62k, Rf2=68k, G= +-23dB
There is an additional 47k resistor from +IN to -IN of the non-inv, to get equal noise gains for both halves.
One might scale frequencies up , using 22uF (~x2.1) caps. Those caps have quite big tolerance anyway...
I dare to repeat, Scott: go for the parallel variation...
Regards, Klaus
@Andrew: Like the old saying goes "two engineers will have five opinions", I see some things a little different...
There is only one input filter Cin, Rin which is effective only for the non-inv, and two filters in the feedbacks. The feedback filters should be matched, while the input filter should not add additional roll-off and phase. Exact matching is impossible, though, because the non-inv drops to gain=1 for DC while the inv drops to zero. In the extreme case of very little gain, say 2dB, there is no -3dB roll-off to be reached for the non inv.
There is a big gotcha in the schem, because if the input is left open, the ac noise gain of the inv is 68k/(47k+4.7k)=1.3 which quite certainly causes the inv to oscillate, given the phase margin of the chip. A buffer is necessary, therefore.
After a little calculating and simulating I arrived to the following values:
Rin=47k, Cin=47uF ==> Fc=0.072Hz, This RC is before the buffer, of course.
Ri1=Ri1=4.7k, Ci1=Ci2=47uF ==> Fc=~0.72Hz
Rf1=62k, Rf2=68k, G= +-23dB
There is an additional 47k resistor from +IN to -IN of the non-inv, to get equal noise gains for both halves.
One might scale frequencies up , using 22uF (~x2.1) caps. Those caps have quite big tolerance anyway...
I dare to repeat, Scott: go for the parallel variation...
Regards, Klaus
Attachments
I originally came up with a different conclusion.
I have read and re-read this comment.
I still disagree, convince me.
The non-inverting is clearly filtered, a series cap Cin feeding a resistor Rin to ground is a high pass filter.
The inverting input is also a series cap Ci1 feeding a resistor Ri2 to ground (this time a virtual ground at the inverting input pin). That high pass filtering action is no different to the non-inverting. I think these input filters should be matched so these set the bandwidth limit.
The fact that the inverting also uses the same string of components Ci2,Ri2 to set the stage gain is not important to the filtering action.
Next, why should the NFB filters be matched between the inverting and non inverting stages? I note your graph shows the mismatch but it only becomes significant at well below 1Hz.
I note you are the third person to support ditching National's version of bridged and advising going parallel.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.