In this thread, I will discuss a concept I find interesting and promising: tandem regulators.
The concept was introduced in this discussion, but it was quickly drowned by a thousand other topics, and I think it deserves better.
We start with a regular shunt reg (first pic). The CCS and shunt regs are easily identifiable.
The power supply rejection is good, a few hundreds µV residual, but the output impedance is less satisfactory (second pic, output current 7.5mA, 5mA pp variation).
Now, let's add a few components (third pic): we can check that they have negligible effect on the initial performances.
Now, let's establish the link between the two regulators (4th pic). The PSR is marginally degraded, but the effect is not significant.
By contrast, the effect on the output impedance is quite dramatic (5th pic).
Another noteworthy effect is that the output current can be increased beyond 25mA without losing the regulation (6th pic)
The concept was introduced in this discussion, but it was quickly drowned by a thousand other topics, and I think it deserves better.
- Tandem regulators in a nutshell:
A shunt regulator normally comprises two elements: the shunt part itself, and a constant-current source (or equivalent function).
Ordinarily, these two element are completely independent; this means that the CCS has to be dimensioned for the maximum output current the regulator will ever see (+ a regulation margin), and the shunt element will see the full variation of the output current.
This leads to high dissipation and poor output impedance.
In a tandem combo, the CCS has its current coupled to the shunt current: this means that variations in the output current will be mostly compensated by an equivalent variation of the CCS.
The advantages are pretty obvious in terms of dissipation and output impedance.
Note that the regulator retains its shunt nature: the shunt regulation is the fast, dominant loop, the CCS being just a "helper": if the link fails, the regulator reverts to the ordinary shunt mode.
We start with a regular shunt reg (first pic). The CCS and shunt regs are easily identifiable.
The power supply rejection is good, a few hundreds µV residual, but the output impedance is less satisfactory (second pic, output current 7.5mA, 5mA pp variation).
Now, let's add a few components (third pic): we can check that they have negligible effect on the initial performances.
Now, let's establish the link between the two regulators (4th pic). The PSR is marginally degraded, but the effect is not significant.
By contrast, the effect on the output impedance is quite dramatic (5th pic).
Another noteworthy effect is that the output current can be increased beyond 25mA without losing the regulation (6th pic)
Attachments
I will also use it, but the large signal behavior provides a more reliable insight into the circuit performance.
Many implementations of the principle are possible, obviously: I have already shown a few different ones, and I am just beginning to explore the concept.
An interesting thing is that most of these implementations lend themselves to further improvements, in particular compensation of residual errors.
Here is an example: it is a higher power, more elaborate version, using 4 transistors.
I am not too confident about an intrinsic superiority of this particular circuit, but it allows an easy insertion of compensating signals.
The two main problems we are contending with are the PSR and the output impedance.
We have seen that the PSR hadn't been improved with the didactic example I had provided (it had been slightly degraded in fact).
This is obviously the first thing to fix.
Imperfect PSR is caused mainly by the dynamic impedances of junctions, like D2 D3, and the Early effect, in Q3 for example; basically resistive effects for the first order. This means that the (first order) compensation is as simple as adding one resistor, R7 in this example.
As a result, the PSR remains below -107dB for the whole audio range.
The output impedance is already good, but it can also benefit from the same kind of trick: instead of returning R1 directly to the ground, it is connected upstream of a current-sensing resistor, R6.
This simple fix (one additional resistor) lowers the output impedance to <200µΩ for the whole audio range.
Some important things are worth noting: at this stage, no capacitor is present (look mum, no hands!), and no additional feedback is involved: just plain feedforward effects, thus no loop stability issues, etc.
Now, a word about the issue raised by Dadod: things look very nice in the frequency domain; too nice perhaps: this type of circuits is supposed to work for a wide dynamic range, it is one of the selling points, compared to the fixed-current shunt regulator.
Let's examine what happens in the time domain: we see at once that the situation is more complicated: the compensation only works ideally for part of the current range, resulting in an undercompensation above, and overcompensation below.
For the large signal conditions, a better compensation is achieved by making R6=340mΩ instead of 300mΩ.
In this case, the circuit becomes a near perfect frequency doubler (albeit a very low level one: the p-to-p level is about 35µV, for a current excursion ratio of 1:3.
However, it shows that when you arrive at this kind of level, a purely linear analysis doesn't tell the complete story, which is why a transient analysis is a good reality-check
An interesting thing is that most of these implementations lend themselves to further improvements, in particular compensation of residual errors.
Here is an example: it is a higher power, more elaborate version, using 4 transistors.
I am not too confident about an intrinsic superiority of this particular circuit, but it allows an easy insertion of compensating signals.
The two main problems we are contending with are the PSR and the output impedance.
We have seen that the PSR hadn't been improved with the didactic example I had provided (it had been slightly degraded in fact).
This is obviously the first thing to fix.
Imperfect PSR is caused mainly by the dynamic impedances of junctions, like D2 D3, and the Early effect, in Q3 for example; basically resistive effects for the first order. This means that the (first order) compensation is as simple as adding one resistor, R7 in this example.
As a result, the PSR remains below -107dB for the whole audio range.
The output impedance is already good, but it can also benefit from the same kind of trick: instead of returning R1 directly to the ground, it is connected upstream of a current-sensing resistor, R6.
This simple fix (one additional resistor) lowers the output impedance to <200µΩ for the whole audio range.
Some important things are worth noting: at this stage, no capacitor is present (look mum, no hands!), and no additional feedback is involved: just plain feedforward effects, thus no loop stability issues, etc.
Now, a word about the issue raised by Dadod: things look very nice in the frequency domain; too nice perhaps: this type of circuits is supposed to work for a wide dynamic range, it is one of the selling points, compared to the fixed-current shunt regulator.
Let's examine what happens in the time domain: we see at once that the situation is more complicated: the compensation only works ideally for part of the current range, resulting in an undercompensation above, and overcompensation below.
For the large signal conditions, a better compensation is achieved by making R6=340mΩ instead of 300mΩ.
In this case, the circuit becomes a near perfect frequency doubler (albeit a very low level one: the p-to-p level is about 35µV, for a current excursion ratio of 1:3.
However, it shows that when you arrive at this kind of level, a purely linear analysis doesn't tell the complete story, which is why a transient analysis is a good reality-check
Attachments
Last edited:
At this stage, this type of practical detail doesn't concern me: as soon as a normal OP bypass cap is added (100µF for example), the peak will be polished off.
Well not completely, but as I said, at this stage this kind of practical detail doesn't really matter.
When the time comes to actually dimension and optimize a circuit that is going to built and tested, this kind of issue will be adressed.
For the moment, I explore principles and topological alternatives.
Attachments
OK, I am waiting for real life implementation, it's very interesting. I found you out of box thinker, one of very few here.
Hi, Elvee,
The concern I would have about the tandem approach is that, at least in principle, it's fundamental operation works to reduce the primary benefit of using a shunt regulator in the first place. Which is the localization of the A.C. current demanded by the load circuit, and also presenting a constant load then to the supply outside the regulator.
The concern I would have about the tandem approach is that, at least in principle, it's fundamental operation works to reduce the primary benefit of using a shunt regulator in the first place. Which is the localization of the A.C. current demanded by the load circuit, and also presenting a constant load then to the supply outside the regulator.
Last edited:
Hi Elvee,
i like the idea and the simplicity of the didac example implementation. I would like to give this a try for a phono preamp i´m working on.
swapping the LED an the two 1n4148 seems to improve PSRR as well as load regulation a little bit, any concerns? is there something i'm missing with this modification?
please allow me a second question: what would be a good starting point for the "quiescent current" (i.e. without load) for such an application? nothing fancy, that 1510 and lm4562 per channel, according datasheet 8mA and 12mA quiescent supply current, approx 40mA for two channels without signal.
thanks very much in advance
thomas
i like the idea and the simplicity of the didac example implementation. I would like to give this a try for a phono preamp i´m working on.
swapping the LED an the two 1n4148 seems to improve PSRR as well as load regulation a little bit, any concerns? is there something i'm missing with this modification?
please allow me a second question: what would be a good starting point for the "quiescent current" (i.e. without load) for such an application? nothing fancy, that 1510 and lm4562 per channel, according datasheet 8mA and 12mA quiescent supply current, approx 40mA for two channels without signal.
thanks very much in advance
thomas
If it seems to work better, there is no reason not to use it.
For such a small output, the quiescent current can be substantial, like 50% of the load. For higher currents, ~10% of the load seems reasonable and will still give good results.
Even a small current will give an improvement, thanks to the synergy
Last edited:
I don't think so, the large 12V obscures the ripple performance.
You should subtract the 12.74VDC from the output voltage to show the ripple.
BTW Your circuit now looks very much like a SE DC amplifier ;-)
Jan
The small-signal, AC analysis shows the idealized performance for a single voltage/current coordinate. Regulators need to cope with a variety of dynamic conditions, and their performance will always be inferior under those conditions, and they may add non-linear components to the output impedance. Such quirks will remain invisible to the AC analysis (but not to a difficult real-world load).
Both types of tests are useful and provide some insight into the behaviour of a circuit.
If you test a regular shunt regulator with a load that exceeds its standing current for a part of the cycle, it will crumble for that part. With a tandem regulator, this will not happen: the series member will rescue the shunt section and provide whatever current is needed.
Such a behaviour is completely unnoticeable with AC analysis, which makes one of the main benefits of the topology invisible.
The regulator may look like a SE amplifier, and it is in fact an amplifier,(look here: https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7074596), but not quite SE: both halves do an active work.
An example of SE amplifier would be a one-quadrant series regulator loaded by a current sink, like this example: https://www.diyaudio.com/community/threads/se-class-a-regulator-chip-amp-madness.192934/post-2644595
This one probably falls outside of the category: https://www.diyaudio.com/community/...the-complementary-version.193214/post-2649553
Both types of tests are useful and provide some insight into the behaviour of a circuit.
If you test a regular shunt regulator with a load that exceeds its standing current for a part of the cycle, it will crumble for that part. With a tandem regulator, this will not happen: the series member will rescue the shunt section and provide whatever current is needed.
Such a behaviour is completely unnoticeable with AC analysis, which makes one of the main benefits of the topology invisible.
The regulator may look like a SE amplifier, and it is in fact an amplifier,(look here: https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7074596), but not quite SE: both halves do an active work.
An example of SE amplifier would be a one-quadrant series regulator loaded by a current sink, like this example: https://www.diyaudio.com/community/threads/se-class-a-regulator-chip-amp-madness.192934/post-2644595
This one probably falls outside of the category: https://www.diyaudio.com/community/...the-complementary-version.193214/post-2649553