Memory Distortion? and some new beginnings.

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Having built two new amplifier modules that originally started out as doug selfs blameless have gradually grown into something that has created some distance from its predecessor. I have to say, I am very excited abut the Current Feedback pair differential as described in the memory distortion articles, I have built it using 2n5551/5401 biased to 4.5ma tail current. It sounds quite neutral, wonderful, lots of seperation and lucidity. Now I am just curious, has anyone else put this configuration to test?. I had to make some compensation changes, upping the lag comp to 30pf and lead to 5pf to maintain stability. drivers were changed from mje15032/33 to 2sc4793/2sa1837..Any opinions on tail current is 4.5ma neccessary or overkill?.


Colin
 
Thank you Colin for restarting this thread... I have been haveing great fun with simulations etc since someone posted those Perrot links....

It struck me how much attention people spend on avoiding distortion mechanisms, yet when it comes to heat induced distortion, they are happy slapping a heatsink on and congratulateing themselves for not reaching thermal meltdown....

Yet they pay attantion to even RF from bridge diodes etc...
However these heated components present a small voltage while cooling down and if I understand correctly useing ohms law, when this meets a high impendance stage you are sitting with a potentialy nasty distortion mechanism, unless you swamp it with feedback...
 
Early Effect

Hi Colin, Nordic,

Let's see if we can avoid another blistering row in this thread!!

I'm not so sure the distortion is all thermal. It's perhaps more related to variation of base/emitter voltage with collector current.

In a conventional bipolar transistor, as you increase collector current the base emitter voltage increase, non-linearly, according to Early effect. This is pronounced with long tailed pairs, and leads to an S shaped transfer function, not unlike the BH curve which defines the magnetisation flux of a transformer.

This S shaped curve means that transconductance of a LTP steadily reduces as an increasing differential voltage is applied to both inputs. The differential input to a power amp is higher than one thinks; if the feedback factor of an amp is 1000, for example (60dB of feedback), then the signal differential between input and feedback nodes is 80mV at 100W/8R output (80Vpp). This transconductance compression is of course dealt with by the feedback network, but it is not eliminated, only reduced.

If you can run the LTP devices in constant current, the Early effect is eliminated, and the transfer function is essentially linear, viz straight.

Colin has already noticed the sonic differences......

Cheers,

Hugh
 
Re: Early Effect

AKSA said:
If you can run the LTP devices in constant current, the Early effect is eliminated, and the transfer function is essentially linear, viz straight.
Cheers,

Hugh

Hello Hugh,

Running LTP in constant current means replacing the emitter resistor with Constant current source......results in linear transfer curve then did you use resistor in your amps...

Plz correct me if I were wrong...

regards,
Kanwar
 
How do you feel about the writer's opinion, that constant power. i.e. constant voltage AND constant current, would lead to even lower thermal load...

I think some of the claimed improvements are that the peak temperature produced by any one cycle of the signal is shared between more devices, leading to A smaller increase in temp per device, leading to quicker return to ambient temperature...

Once a peak is delivered the device is heated... the problem is that long after the signal passed, the device is not at the same thermal point it was just before the signal reached it... Like you can quickly boil a kettle of water but it will take much longer for it to cool back to ambient if just left to stand...
 
Nordic,

Constant power is a great idea.

However, consider the electrical environment. If we can achieve constant current, just how much does voltage vary anyway?

Assume a 36V rail, and a 2Vpp signal. This means power might vary from (37 x 1mA) to (35 x 1mA), that is 37mW down to 35mW, with a mean around 36mW. This is plus and minus 2.78%. This is not a large variation, and will not, in my view, result in much variation of temperature on the semi die, particularly at frequencies over 50Hz.

Cheers,

Hugh
 
http://peufeu.free.fr/audio/memory/memory-1-theory.html

The black curves show the drift of input pair error voltage Vb(Q3)-Vb(Q4) ; the bottom one is a closeup :

At T=0, VError = 4.98mV
During the impulse, for an output of 30V, VError = 9.302mV (This gives a rather lowish openloop gain of 6940)
Immediately after the transient, VError = 5.177mV
The thermal drifts therefore caused VError to shift of 197uV (this corresponds to the 1.37 value at the output if we multiply it by the OL gain). If we compare this to the VError during the impulse (4.322mV), we come to this shocking conclusion :

The thermal drift signal is only 27dB down from the musical signal !

other pages
http://peufeu.free.fr/audio/memory/

input-evolution.gif


The memory in the buffer transistor is quite negligible compared to the VAS transistor. Imagine the VAS is confronted to a 50k load (base resistance of the buffer), with a 10mA current sink. Heating the VAS transistor by one degree C will vary its Beta by 3%, which means a drift of 10mA * 3% * 50k = 15 volts that will have to be corrected by feedback. Heating the buffer by 1°C will only make it drift 2mV... The essence of the thing is that memory is multiplied by the gain of the stages downstream ; the VAS has a lot of gain, the follower has none.

If you use an emitter follower for amplification (with for instance 10 Ohm from emitter to ground and 1k from collector to V+) as is done in an input diff pair, you're into trouble, because the Vbe variations are amplified by the gain of the stage. They have to be compared to the input signal, not the output signal.

With a buffer and a small output stage, it is actually an opamp, even if a complicated one... It could be used for anything actually.
 
This really sounds good!

Maybe I should start my own VA-Amp thread, lol, well maybe not. To Nordic, I considered posting under the Compensation thread but since this is unrelated I thought it warranted a new start to things or a new thread. I am absolutely fascinated by the MD articles, and the Lavardin patents as this seems to go a little farther than what I can find from Self considering input stages. Most input stages are much the same blood curent mirro/no current miror, more,less,no degeneration, ccs/resistor seems to be the only thing that has seperated them from the others, this concept struck me as different enough to warrant a complete re-designed, and new pcb layout to utilize, alot of work when
drawing out by hand and having to make two boards :). I have to thank Hugh on this one, unless you google Gerard, or Memory Distortion(a concept I would not have thought of overnight)its almost impossible to find these pages!. I found Mike B;s post on this input in the archive, he seemed to have much the same reaction to it as I have.


Colin
 
http://herkules.oulu.fi/isbn9514265149/html/x781.html
Saw this paper?

Thermal impedance describes the relationship between dissipated power and temperature, and block K describes the relationship between temperature and the gain of the amplifier. Only the gain of the amplifier is considered to be temperature-dependent for the present purposes. In practice, however, output conductance (Fox et al.1993) and capacitances are also temperature-dependent, as will be seen in Chapter 4. Some of the circuit parameters of the transistor are always functions of temperature, and since temperature-compensated external bias networks do not detect the junction temperature, they are far too slow to compensate for its effects. Thus, no improvement can be expected in terms of TPF.
 
At the risk of sounding like a "resistor value changes everything" sort of guy, it seems to be a key element in the best sound in this pair. In configuration B R5 and R6 I settled on 1.5k, higher values seemed to give a harder edge to the highs now all is silky, lovely!, noise is lower too..I am using 100ohm degeneration on the emitters.


Back to enjoying sweet sweet music.

Colin
 
Ah man, you mention resistor values and then the thread dies an abrupt death? :). Still listening to music, cannot believe the tonal contrast that was missed with a standard differential, ambience, spatial cues and most of all the lack of any detectable hi-frequency distortion!..



Colin
 
Colin,

I run stage current of 3.6mA, R5/R6 of 1K5 like you, and 2 x 33R for degeneration. This means around 0.4mA in each LTP device, and 1.4mA in each EF. Making the CCS adjustable allows you to accurately fix offset.

If you are taking the output off one side, ie single ended, then the two EF resistors to rail should be slightly unequal to promote current balance, taking account of the 60uA or so of bias required to drive the VAS base.

Cheers,

Hugh
 
Hello Hugh, I have been running the tail current at 4.5ma with 100 ohm degeneration something still engrained from the blameless. I am using a current mirror collector load with 49.9 ohm emitter degeneration resistors being run single ended taking the output to VAS from the input ltp collector between mirror and r5., I am wondering if I should try the resistor to rail instead?. I guess loweringf the stage current I should be able to also lower the degeneration resistors, At 3.6ma noise should be lower, but will Slew Rate not also suffer slightly?.


Thanks
Colin
 
Hi Andrew,

Getting a handle on SR comes back to level, like rise time. If we work on the 3dB point, where I measure 82.5KHz, then SR is Vp-p x 2 x KHz/1000. For 80Vpp into 8R, or 100W, this is 13.2V/uS.

If you measure the 6dB point, it's a lot lower, around 10V/uS.

In any event, it's more than sufficient to frighten six months growth out of a bat.......

Cheers,

Hugh
 
This is an interesting topic, but I am afraid the thread has started out with a few misunderstandings, that are better cleared up for the benefit of the discussion. I am sorry it got much more lengthy than intended, but I also discuss the actual topic a bit.


vynuhl.addict said:
I have to say, I am very excited abut the Current Feedback pair differential as described in the memory distortion articles

Presumable you are referring to the topology in figure B, posted by Nordic. The correct term is complementary feedback pair (CFP) or Sziklay pair, not current feedback pair. And to avoid myself causing further misunderstanding, figure B shows a differential pair (or LTP if you wish) with each of the usual two transistors replaced by a CFP. It is not the whole thing that is a CFP.


AKSA said:
In a conventional bipolar transistor, as you increase collector current the base emitter voltage increase, non-linearly, according to Early effect.
[/B]

I am afraid not. You are quite right that collector current and base emitter voltage are related, more precisely they are exponentially related in the first order model. This is the most fundamental property of the BJT. However, this is not the Early effect. The Early effect is the phenomenon that if the base current is held constant, the collector current will vary with the collector emitter voltage (due to base width modulation). So both phenomenae exist, you just got the terminology wrong. (Blame it on a bad day Hugh, I know you know this.)


AKSA said:
This is pronounced with long tailed pairs, and leads to an S shaped transfer function, not unlike the BH curve which defines the magnetisation flux of a transformer.
[/B]

Yes, while a single BJT has an exponential relationship between Vbe and Ic, a diff pair has an S-shaped relationship between the differential base voltage and each of the collector currents. This is both good and bad. The good thing is that for very small differential voltages, a diff pair is much more linear than a single BJT, which is always exponential. However, as the voltage gets bigger, the diff pair quickly gets more nonlinear than the single BJT, so it is important to try avoiding big differential voltages.

(As a sidenote, this is one of the potential causes of TIM, or whatever we might call it, that due to slewrate limitations, insufficient frequency response etc. the feedback signal is not "fast" enough to keep up with the input signal. As a result, a transient on the input might cause a transient differential voltage that goes far outside the "linear" region of the diff pair. As Leach points out, the input signal should be LP filtered and the amplifier fast enough so this never happens.)


AKSA said:
If you can run the LTP devices in constant current, the Early effect is eliminated, and the transfer function is essentially linear, viz straight.
[/B]

Sorry again, Hugh. I assume you mean keeping the tail current as constant as possible, eg. by using a CCS. As said before, this has nothing to do with the Early effect. However, it also does not essentially change the transfer function of the diff pair, which will remain S-shaped. Rather it will be a more perfect hyperbolic function than if you use for instance a resistor instead of a CCS. The S-shape of the diff pair is inherent to it. Making the tail current more constant impoves the CMRR, which also reduces second-order errors in the transfer function,



Nordic said:
How do you feel about the writer's opinion, that constant power. i.e. constant voltage AND constant current, would lead to even lower thermal load...

First, I hope I am not the only one finding the terms "constant current" and "constant power" inappropriate here. While cascoding, as shown in fig. C, is a method to attempt achieveing a constant Vce of the transistor (and the type of cascoding used in this case is probably as good as it can get without resorting to much more complex solutions), a CFP diff pair has nothing to do with constant current, in my opinion.

The idea of the CFP is as follows. A BJT has an exponential transfer characteristic, as mentioned above. This means that the relative error gets larger further away from the Q point, that is, it gets more nonlinear the larger the current swing. To improve the linearity of a BJT, one of the best things one can do is to keep the current swing on the collector as small as possible relative to the Q current. Often we cannot achieve this, since it would require an unreasonable Q current. The CFP attempts to fix this problem. In a well designed CFP (at least for small signal purposes) the input transistor of the pair has a very small current swing and the second transistor acts as a current booster, which provides almost all of the required current swing. Further, the input transistor should preferrably still have a fairly large Q current despite its small current swing. Then the input transistor will be much more linear than a single BJT would have been. The booster transistor, however, does not get that advantage, since it has to provide the current swing, but this transistor is inside a local feedback loop within the transistor pair, which compensates for most of its error. Hence, it is essentially the error of the input transistor that matters, and this error has been considerably reduced. Generally, a CFP will have lower THD, but with more pronounced high order harmonics in its spectrum, than will a single BJT. So, this has nothing to do with constant current, but with minimizing the current (swing). Saying that it is so low that it is almost constant is in my opinion very misleading.

So here we arrive at what Hugh said:

AKSA said:

I'm not so sure the distortion is all thermal. It's perhaps more related to variation of base/emitter voltage with collector current.

As I described above, using CFPs instead of single BJTs, we improve the (electric) linearity of the diff pair considerably. That means that even if there is no thermal effects at all, we should expect improved performance (speaking of measurements, to avoid the discussion of audibility etc.).

Hower, this does not rule out the possibility that thermal effects matter too. As I said above, the booster transistor is within a feedback loop and it is mostly the error of the input transistor that matters. Since we have reduced the current swing considerably in this transistor, we have also considerably reduced the power variations, and thus the temperature variations in it. So regardless of whether thermal effects was a problem in the ordinary diff pair, it should be even less of a problem with the CFP diff pair, and cascoding will have a further effect in reducing the power variations.

The question remains, how much of the improvement is due to electrical issues and how much is due to thermal issues? Unfortunately, it seems very difficult to measure the thermal effects since it is the temperature variations of the junction that matters. It is probably easier to make calculations of the expected effect using a good thermal model of the devices, as shown in one of the links. Unfortunately, that example used a TO126 package, which is not so interesting for input stage diff pairs. Finding a thermal model with thermal impedances, not only resistances, for a TO92 package would be helpful here. Does anybody have such a model?

Finally it should be pointed out that CFP diff pairs is not unkonwn elsewhere in the literature. I is mentioned by Self and many others, for instance, although they ususually only discuss the electrical properties, not the thermal ones. CFP diff pairs can also occasionally be found in commercial amps, eg. Halcro.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.