gengcard said:I have two question for T-network. Is it able to be applied to OPA549? I saw several people applied T-network to only LM3875. Why has nobody try with OPA549?
Because a number of us use various forms of buffers including tubes in combination with inverted 3875. This thread is an offshoot of an earlier one that deals with buffered 3875s. Does it apply to OPA549? Why don't you try it? That way you can tell us. Will it work? Very likely. Will it improve the sound? Very likely. But are you using 549 inverted or non-inverted? Here we are talking inverted 3875.
Joe R.
As I mentioned in an earlier posting: the OPA549 will not be the best candidate for a t-network, because of his limited gain/bandwith product.
Why not a quick, reversable test?
But: not just the LM3875. For inverted configurations the t-network will give you a better sounding amp even with LM3876 and LM3886 imho.
Franz
Why not a quick, reversable test?
But: not just the LM3875. For inverted configurations the t-network will give you a better sounding amp even with LM3876 and LM3886 imho.
Franz
bigparsnip said:
Can you elaborate a little bit your statement, please? (perhaps with a sketch or a diagram).
I don't agree (I don't understand, but it's all my fault :-()
Just to set a common language, I speak about a voltage amplifier (not a transimpedance one) with or without T network and I study the circuit starting from his (so called) "noise gain"
"Life is like a box of chocolate, you never know what you're going to get." Forest Gump
One thing is to speculate about things, but it is better to find the answer in the real world.
Example One:
Blue is before conventional feedback, Red is T-network, reference 2.83V into 8 Ohm = 1W
You can make out that Red averaged out is lower than Blue. In fact, running a marker over the traces indicate Red better by 3dB approx. T-network wins the round.
Example Two:
Blue is before conventional feedback, Red is T-network, reference 1V RMS into 8 Ohm = 125mW
Here the difference is even more obvious. Note too that the noise doesn't rise with frequency - T-network is clearly less noisy.
These are not simulations, they are actual measurements.
You can of course also compare harmonic distortion. Does it also look a tad lower to you?
Joe R.
One thing is to speculate about things, but it is better to find the answer in the real world.
Example One:
Blue is before conventional feedback, Red is T-network, reference 2.83V into 8 Ohm = 1W
An externally hosted image should be here but it was not working when we last tested it.
You can make out that Red averaged out is lower than Blue. In fact, running a marker over the traces indicate Red better by 3dB approx. T-network wins the round.
Example Two:
Blue is before conventional feedback, Red is T-network, reference 1V RMS into 8 Ohm = 125mW
An externally hosted image should be here but it was not working when we last tested it.
Here the difference is even more obvious. Note too that the noise doesn't rise with frequency - T-network is clearly less noisy.
These are not simulations, they are actual measurements.
You can of course also compare harmonic distortion. Does it also look a tad lower to you?
Joe R.
Konnichiwa,
The T-network is a long understood and well known feedback circuit, it is found in > 50 Year old Valve gear.
The key advantage is that for a given gain you are free to select any value of feedback network desired for your application and thus the input impedance can be selected freely, with a passing nod to noise performance which is invariably worse.
HOWEVER, there is one KEY problem with the T-Network.
Namely DC Gain. With the T-Network the DC gain is high, unless a large value blocking capacitor is inserted in series with the resistor to ground in the T-Network. The classic shunt feedback network has the input DC Blocker (which is usually neccesary) and nothing else. The T-Network requires the addition of a further capacitor in order to give the same DC performance.
The final question is another. Namely, are "High" value resistors in the feedback circuit a problem?
The answer is that this depends upon the circuit inside the Op-Amp Chip. In the case of the LM3875 the inputs are bipolar with an added emitter follower, making the input current modulation small but finitly small. As a result the non-inverting input requires carefull impedance matching to get things right in the case a high value feedback resistor is used.
So, a low impedance feedback network may very well be desirable. One must however consider the cost paid in either having a feedback loop DC blocker and/or excessive DC offset.
The solution to the offset is of course a DC servo which CAN be implemented in such a manner that it's operation is very low in audibility by making sure the "pull-range" of the servo is limited to the absolute minimum acceptable.
However we look at it, we invariably trade one problem for another.
As DIY'er I would be very well tempted to solve the problem with a 100R/2K2 feedback network and a BUF 634 in wide bandwidth mode and to include a 2nd order non-inverting servo to the biasing for the BUF634 input with the only coupling capacitor placed at the input of the BUF634 with a 100K/0.47uF combo.
But then, I would probably also be tempted to make the circuit differential (balanced) to use crosscoupled error correction and many other things. It all depends upon your design goals.
Sayonara
Franz G said:
The T-network is a long understood and well known feedback circuit, it is found in > 50 Year old Valve gear.
Franz G said:
The key advantage is that for a given gain you are free to select any value of feedback network desired for your application and thus the input impedance can be selected freely, with a passing nod to noise performance which is invariably worse.
HOWEVER, there is one KEY problem with the T-Network.
Namely DC Gain. With the T-Network the DC gain is high, unless a large value blocking capacitor is inserted in series with the resistor to ground in the T-Network. The classic shunt feedback network has the input DC Blocker (which is usually neccesary) and nothing else. The T-Network requires the addition of a further capacitor in order to give the same DC performance.
The final question is another. Namely, are "High" value resistors in the feedback circuit a problem?
The answer is that this depends upon the circuit inside the Op-Amp Chip. In the case of the LM3875 the inputs are bipolar with an added emitter follower, making the input current modulation small but finitly small. As a result the non-inverting input requires carefull impedance matching to get things right in the case a high value feedback resistor is used.
So, a low impedance feedback network may very well be desirable. One must however consider the cost paid in either having a feedback loop DC blocker and/or excessive DC offset.
The solution to the offset is of course a DC servo which CAN be implemented in such a manner that it's operation is very low in audibility by making sure the "pull-range" of the servo is limited to the absolute minimum acceptable.
However we look at it, we invariably trade one problem for another.
As DIY'er I would be very well tempted to solve the problem with a 100R/2K2 feedback network and a BUF 634 in wide bandwidth mode and to include a 2nd order non-inverting servo to the biasing for the BUF634 input with the only coupling capacitor placed at the input of the BUF634 with a 100K/0.47uF combo.
But then, I would probably also be tempted to make the circuit differential (balanced) to use crosscoupled error correction and many other things. It all depends upon your design goals.
Sayonara
Konnichiwa,
Well, you could use a 4u7/10K/1M Circuit in the NFB loop, but yes, the whole thing is, as said, trading one problem for another.
I would probably recommend a servo. The trick is to insert a voltage divider between the Servo Op-Amp output and the servoed input of the Op-Amp.
For example, try (no load attached) how many mV on the positive input deflect the actual Chip-Amplifier output by around +/-1V. With a DC Gain of ~ 100 I would expect around +/-10mV to do the trick. You can then set the servo which will probably allow a +/-12V output with a +/-15V supply to produce the said 10mV when it's output is at 12V.
In other words the AC (noise, distorted signla etc) is reduced from 12V "full scale" to 12mV "full scale" or in other words any sonic impact of the servo is reduced by 60db compared to the usual misapplication of having a by far too wide pull-range by connecting the servo output pretty directly into the audio circuit. Combine that with a decent servo op-amp and things should go great and will likely sound better than a big 'lytic.
Sayonara
Franz G said:
Well, you could use a 4u7/10K/1M Circuit in the NFB loop, but yes, the whole thing is, as said, trading one problem for another.
I would probably recommend a servo. The trick is to insert a voltage divider between the Servo Op-Amp output and the servoed input of the Op-Amp.
For example, try (no load attached) how many mV on the positive input deflect the actual Chip-Amplifier output by around +/-1V. With a DC Gain of ~ 100 I would expect around +/-10mV to do the trick. You can then set the servo which will probably allow a +/-12V output with a +/-15V supply to produce the said 10mV when it's output is at 12V.
In other words the AC (noise, distorted signla etc) is reduced from 12V "full scale" to 12mV "full scale" or in other words any sonic impact of the servo is reduced by 60db compared to the usual misapplication of having a by far too wide pull-range by connecting the servo output pretty directly into the audio circuit. Combine that with a decent servo op-amp and things should go great and will likely sound better than a big 'lytic.
Sayonara
Re: Pre Amp
Here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=40783
Not yet, it's still in my head and in the prototype.
Audiofanatic said:
Here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=40783
Audiofanatic said:
Not yet, it's still in my head and in the prototype.
Hi there,
I have been following the posts about DC offset with interest. I noticed a higher DC component after the implementation of the T-network, but it dropped with a resistor to the non-inverting input. These measurements where made at idle. Am I to understand that the DC is created when the amp is been driven, and not when at idle. If so has anyone actually observed it as a problem ?? How much DC offset are we talking about ???
Help a poor soul to understand.
Shoog
I have been following the posts about DC offset with interest. I noticed a higher DC component after the implementation of the T-network, but it dropped with a resistor to the non-inverting input. These measurements where made at idle. Am I to understand that the DC is created when the amp is been driven, and not when at idle. If so has anyone actually observed it as a problem ?? How much DC offset are we talking about ???
Help a poor soul to understand.
Shoog
Joe Rasmussen said:
Hi Greg
http://www.diyaudio.com/forums/attachment.php?s=&postid=473584
Sorry, but no, it potentially gives you worse DC offsets (note the plural).
The problem is that your 'reasonable' 47K has been tried and gives rises to DC shifts due to chip input imbalances. It needs to be a lot lower than that and 10K is a better compromise.
Yes, you can go for the lower DC gain, but (don't you just hate that word) it means in your case increasing the R (R16 in your sch) from the (-) input to the "T" junction. Keeping the DC gain around 100:1 (10K/100R) enables us to lower the that R to 10K which is better able to cope with input offset curents and voltages when they are not close to ideal. They are more often than not, but we need to present a consistent solution.
An externally hosted image should be here but it was not working when we last tested it.
It's not the typical values that are of concern. The Max values means the 3875 is still within spec. NS have covered themselves nicely. Your values would deal very well with the typical, but I have here a couple that are not typical and they will blow out the DC parameters. 47K is too high. Think about it, if this was zero, the input difference in current (the offset) would not make any difference. That's not practical, but it shows it needs to be low value. The larger the value the greater the difference will get magnified by the apparent closed loop gain. It goes then that your DC gain has to go high to 100 match.
I have now done 6 channels (chips) using 22K In/10K-100R-10K/10K on (+), and so far all have been under 50mV. When others try that combination of values, we will be able to find whether we have limited DC Offset to below or around 50mV with a larger sample. The spread that I got below 50mV is significant, ranging from 6mV to 38mV so far. Time will tell whether we have a consistent solution.
Why couldn't they have used fet inputs?Wouldn't sound as good I suspect.
Joe R.
Thanks Joe.
Because the I-offset is the problem, want would be the disadvantage then to still do what I proposed (DC gain of 5-10 and add AC gain with 47k FB res.) and fine-tune the resistor from +in to GND (not injecting voltage) to get the Output DC offset to zero. It sounds like the simplest way to solve the Offset issue. It'll create only a minor Z imbalance which shouldn't be a problem.
/Greg
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.