New Posts New Posts RSS Feed - 1970s Design Indulgence
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

Welcome to the Graham Slee Audio Products Owners Forum

 

Open to all owners plus those contemplating the purchase of a Graham Slee HiFi System Components audio product and wishing to use this forum's loaner program: join here (Rules on posting can be found here)

This website along with trade marks Graham Slee and HiFi System Components are owned by Cadman Enterprises Ltd


1970s Design Indulgence

 Post Reply Post Reply Page  <1 9899100101102 345>
Author
BAK View Drop Down
Senior Member
Senior Member
Avatar

Joined: 14 Mar 2010
Location: Kentucky, USA
Status: Offline
Points: 1744
Post Options Post Options   Thanks (1) Thanks(1)   Quote BAK Quote  Post ReplyReply Direct Link To This Post Posted: 04 Apr 2020 at 1:20am
Originally posted by Graham Slee Graham Slee wrote:


Listening tests however, suggest this is far more palatable because sibilance is vastly reduced, or non-evident, and with electrolytics not having to handle masses of high frequency distortion, the sound with amplifier permanently switched on, has not deteriorated.

The thermal stability is good as witnessed by the almost constant meter readings under quiescent conditions. These can vary upwards to around 9mV, and down to 7mV at room temperatures between 23 and 27 C.

Sounds like great results.
Bruce
AT-14SA, Pickering XV-15, Hana EL, Technics SL-1600MK2, Lautus, Majestic DAC, Technics SH-8055 spectrum analyzer, Eminence Beta8A custom cabs; Proprius & Reflex M or C, Enjoy Life your way!
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 05 May 2020 at 5:49pm
Google MJL21193 negative resistance, and you'll be lucky to find anything to do with it.

I'm publishing the following article for the simple reason, that if I do not, then it will be lost in the annals of time, hidden by the superficial noise of this sad world:

https://www.hifisystemcomponents.com/downloads/articles/Prevent-Emitter-Follower-Oscillation.pdf

After reading it, you will realise why it would be nice to know the MJL21193 negative resistance or any transistor's negative resistance for that matter.

I've had it to the back teeth with all the bollocks often talked about with boastful arrogance, about what makes amplifiers "right," when the simple basics of the above aren't even considered (and that's why articles on such things are so hard to find).
That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 06 May 2020 at 3:17am
Circuits are designed with allowances for production tolerances, just as much as mechanical components.

The stereo output stage, consisting of 2 pairs of transistors, will cost the manufacturer £12 per set. In an EF2 amplifier (which, this is in its current form), each pair are complimentary - one NPN, the other PNP - and although the maker tries very hard to match their characteristics, their differing technology means some things will never match.

The manufacturer has already rejected the dies, which are outside the tolerance, and in British transistor manufacturing's heyday, rejected transistors were often cheaply sold by the ton for hardcore.

Assuming something similar happens in China, what we have left is tubes of 25 plastic TO3 power transistors costing £80, and containing relatively close-tolerance duplicates of each other.

The "proof of the pudding" is revealed on test, and the production audio analyser runs a predetermined test sequence, whose result is either pass or fail.

It is the designer's job to ensure more pass than fail, and to design with what he has available, such that the design has sufficient latitude for the device tolerances. Ideally, the designer has enough manufacturing experience to understand the cost involved in being - shall we say - overly perfectionist.

If an amplifier is to sell at retail for £1200, a manufacturer spending £1200 to do the job, would soon be bankrupt. That should be obvious, but just in case it isn't, running through hundreds of power transistors to get exact matches, costs time as well as component money. I'll let you work that out.

In the next part, some "give" will be included, in which the resulting amplifier is improved, without being so precise as to cost the same as an aero engine.
That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 06 May 2020 at 4:04am
A transistor model contains all the named mathematical variables which make up its characteristics. With the aid of a SPICE program,  all the variables for all the connected devices and passives get computed. It searches component libraries contained in numerous folders, bringing models of each into the calculation. The initial simulation run can take a few seconds as these models are discovered and collected. The actual calculation time is a split second. Doing it by hand calculation would take hours or even days.

You don't need to know what you're doing, as long as you know the shape of the results. You can then play all day, for days or weeks, plugging things in by guesswork, and you might get a result you can build.

But sometimes, you need a greater understanding of the circuit elements and how they interact, which means good old fashioned pen and paper. This can give you valuable insight into what is happening, but you can only really take one circuit section at a time, or you will be bogged down and need a computer program to help you - such as SPICE!

The three kinds of passives: inductors, capacitors, and resistors, otherwise known as LCR, contain small parts of each other, an example being a capacitor having resistance and inductance. Sometimes the designer can ignore these impurities, which result in parasitic effects, but at other times, ignoring them can cause trouble.

The cat's whisker radio gave rise to an understanding that negative resistance could be used to amplify, and although it kind-of failed, the amplifier transistor came into being. Unfortunately, because transistors are impure, and they contain little bits of LCR, they can also exhibit negative resistance, but frequency dependant (FDNR), therefore it should be described as negative impedance.

There seem to be lots going on around FDNR these days, but it is nothing to do with the FDNR in power amplifiers, which, by contrast, is a big 10-77.

FDNR in an emitter follower is the cause of parasitic oscillation, and all amplifiers suffer from it. Some suffer less than others, and those amplifiers tend to sound better.
That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 06 May 2020 at 1:21pm
If negative resistance appeared in parallel to the device, increasing source resistance leads to higher gain.

However, as in valve grid stoppers, series resistance in the base trace to the transistor calms the problem.

Therefore, the negative resistance can be imagined to be a series element. Imagine is the right word because even though there is no such real resistance, it is the combination of trace inductance with parasitic transistor capacitance, as well as the device's base inductance, which results in the negative resistance, which is frequency-dependent.

There is one particular reason why this subject hasn't been broached in detail here before, and it's the "common belief" that EF2's, or output doubles, don't suffer parasitic oscillation. Books lead you to believe it is a problem exclusive to output triples, or more complicated output stages.

Such books fail to give evidence why it should not affect an EF2, and with me not able to conclude the design because of it not sounding the way I want, either no stone is left unturned, or I scrap the project.

There is another point, and that is the continued availability of output devices, still being manufactured, but the news from manufacturers in 2014 stating that production had ceased, obviously means someone else is doing the manufacturing.

Whom could that be? It is obvious it isn't the original manufacturer, but they are still advertising the devices are in production, so they must be being subbed-out.

The only thing I can think of is that counterfeiters have turned "good," picking up where the manufacturer's plant ceased. Certificates of origin indicate China, while data sheets state Singapore. It is a tangled web.

Authors often contradict each other, with one advocating "straight" emitter followers, while another (Self) warns that even a small signal, one transistor emitter follower, will oscillate without precautions. I can vouch for that!

We read in Chessman and Sokel's work that negative resistance can range between 0 and -500 ohms. My take on that is that small-signal transistors might be at the upper end, and power transistors near the lower end.

The thing which justifies such thought is no power amplifier would be much use with higher than 500-ohm "stopper" resistors. Translated to emitter resistance by the factor known as beta, you could have an output impedance of 5 or 10 ohms, which wouldn't be much good in driving an 8-ohm speaker!

However, Chessman and Sokel say such "stopper" resistors "will range from tens of hundreds of ohms," and hopefully that is a misprint, and should say "from tens to hundreds of ohms."

Tens of ohms still represent a problem, as 10 ohms translates to 0.2 ohms of additional output series resistance, and 100 ohms translates to 2 ohms.

What could we do to mitigate this?


Edited by Graham Slee - 06 May 2020 at 3:50pm
That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 08 May 2020 at 12:56pm
The calculation of Vq depends on being able to see or pick out the unseen resistances. Should there be a potential through which current flows, it stands to reason, there is a resistance, for Ohms Law to be fulfilled.

The very fact there is a thermal voltage in series with a transistor emitter, says there is a resistance in there, and by varying the current, that resistance can be made the value you want. However, that voltage is fixed, and no attenuation, save for that which dictates the intrinsic negative feedback of a common emitter amplifier, can occur.

But can we use it as a series resistance in the base lead of the output transistor? Because, if so, we can obtain the "tens to hundreds of ohms" required to prevent parasitic oscillation, to overcome the negative series resistance.

If a real resistor is placed in the base lead, we will 1. attenuate the signal, 2. increase output resistance (lowering DF), and 3. have a DC voltage drop, which might be sufficient to cause a loss in available output power. These three things depend on the size of that resistance.

If much higher than say 2 ohms, we run the risk of the above, although I have seen circuits using 10 ohms. Neither is the "tens" we at least need to overcome the negative resistance.

It is, therefore, worth a try, but to increase the driver emitter resistance requires it to have a lower standing current. That current is set by the emitter to emitter resistor, which is used to sweep the stored charge out of the output transistor capacitance.

By increasing the value of this resistor, means the charge sweeping is less effective, but what have we got to lose? This amplifier does not go all-out for high gain at high frequencies. It cannot because the input/VAS stage does not have the gain or bandwidth of a multiple transistor stage - it only has two.

Charge sweeping is supposed to lower high-frequency distortion, but there is no audible evidence of that - sibilance was still a problem.

Rather than being incremental, I chose a factor of 10, and in fact, just a little over, by making the emitter to emitter driver resistor 3k.

3k is in-fact 2RE, such that each driver RE would be 1k5 in a non-cross-coupled design. I have used 1k5 in the past, and they gave reasonable sound quality.

The voltage drop across the 3k is the 2 base-emitter drops of the output stage, plus a few tens of millivolts for quiescent voltage drop. By calling it something like 1.225V, this resistor's current will be around 0.4mA, instead of the more usual 6mA.

To find the intrinsic emitter resistance due to the thermal voltage (26mV at 25C), we also need the value of current flowing into and out of the output stage bases. Here, we have to hazard a guess at the eventual output stage quiescent current, and we seem to be homing into 35mA.

Taking datasheet DC current gain (Hfe or beta), which doesn't entirely extend leftwards enough for 35mA, we make an educated guess, and we find betas of 55 and 90 respectively. We can then calculate the base current, and once we have, we can add that to the approx. 0.4mA, to find the current through each driver's thermal voltage.

The drawing shows these as being 25 - 32 ohms. Even when in parallel with the 3k cross emitter resistor, the values hardly change. Therefore, we have achieved at least the "tens of ohms" suggested by Chessman and Sokel.

At this point, it occurred to me that these "imaginary" or "invisible" resistances might not be the complete answer, and this is because the voltage drop is fixed at 26mV. However, the ebb and flow of the signal ensures the current changes, and therefore, so does these resistances, and so they become a variable. I, therefore, decided to include a constant, of roughly 1/10th their value, and this is in the form of a real resistor of value 2.2 ohms.

This value is often used in output transistor bases anyway, but is not always seen in an EF2.

I had already placed a lossy ferrite bead on the power transistor base leads, which got me thinking about using carbon films for the 2.2-ohm resistors. Carbon film tends not to be used much in high-class audio because above 1MHz, they're pretty useless (I will not go into the reasons here). The thought occurred that this degree of uselessness isn't an on-off switch. It must, therefore, extend below 1MHz to a lesser degree, and so it might help in overcoming the frequency dependant negative resistance.

As we've arrived at an output stage circuit, where all the values are known, we now need to find how they affect the output resistance, and this needs to be calculated, so we know how to set the quiescent voltages.

This is to be explained next post, and I hope to update the drawing to make the calculations more intuitive.


That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Retired

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 16298
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 08 May 2020 at 7:05pm


The starred resistors are not real physical resistors, but values that exist nonetheless.

The only values we are unable to establish are R- in the upper and lower base circuits. These are the negative resistances, which cannot exist at DC, and are frequency dependant, usually at frequencies beyond 50kHz (according to Chessman and Sokel).

I have not included any lead inductances or any parasitic capacitances. The Rth's of the drivers, and the 2R2 carbon film resistors attempt to overcome these and have been discussed earlier.

The Rth's of the output transistors form no part of the calculations, although they are part of the gm-doubling we wish to avoid. They are shown to depict the thermal voltage of each output transistor, and our task is to ensure the effective emitter resistance does not allow the quiescent voltage to exceed their 26mV.

The effective emitter resistances of each output half are where the quiescent voltages occur, and therefore, the 26mV (and no more) appears across. The Rin/55 and Rin/90, are in fact, whatever the real Rin is, divided by the beta of the transistor. In other words, the Rin values are transferred to the emitter circuits, where they effectively increase the value of the real emitter resistors.

Although the output transistors are complimentary at someplace in their specification, the differences in NPN and PNP physics make it impossible for every characteristic to match. Even using matched pairs, they will only match for one condition, that usually being maximum rated power.

The ac source of current is the VAS collector load, to which the driver bases are attached. The signal swing is across the VAS load, which is pictured to the left of the dotted line.

The bias spreader goes between the driver bases (the trimmer resistor and three series diodes), and "pulls" them apart, to bias the output stage such that the correct quiescent voltage appears at the output.

The actual load is the 1k2 and 3k resistors, plus the ac input resistances of the drivers, which appear in parallel.

Without the bootstrap capacitor, the driver bases would limit the positive-going signal swing. So the output allows positive feedback via the bootstrap capacitor, to lift the upper end of the 3k VAS resistor.

Because the voltage gain of the output stage is less than unity (<1), the bootstrap does not cause oscillation. Therefore, positive feedback might not be the correct terminology but suffices to explain here.

There is still the load of the drivers, calculated here as being 2.7k and 2k, respectively, and these appear in parallel on the lower end of the VAS 3k. Attenuation is avoided to a considerable degree by the bootstrap pushing the upper end upwards.

However, two things should be observed. 1. the output stage gain is less than unity, and so the voltage across the 3k (and hence the current) cannot remain constant - it will dip slightly. 2. it is placing signal swing on the upper end of the 3k "after the event," and so any signal aberration mixes amongst this almost constant voltage (a "constant current" source).

It is best, therefore, to make the load on the VAS as light as possible, but if the drivers were to use the usual 100-ohm emitter resistance, that load would be considerable.

As such, the choice to increase the resistance of the driver's thermal voltages, in trying to swamp the output transistor's negative resistances, has also helped the VAS stage.

There is still sufficient overall beta to translate the 6 or 7 mA of VAS current to that required to drive the loudspeaker - little of it lost by the driver's emitter resistor.

So now, to the calculation, which is quite easy. We wish to see 26mV across each output transistor's emitter resistance (at quiescent/standing current), or 52mV emitter to emitter.

However, each emitter resistance is not the real resistor value that we may place on the board, but a combination of that plus the translated Rin and Rbase totals, divided by the transistor's beta.

Assuming the standing current to be the 35mA (approx.), which gave the best spread of distortion harmonics on previous analyser tests, and knowing the voltage drop should be 26mV, the actual emitter resistance is revealed by Ohms law.

The effective emitter resistance is, therefore, 26/35 = 0.743 ohms.

We have found the base resistances and translated them to emitter resistances. So these are taken away from 0.743 ohms, to give the actual value of physical emitter resistance to use.

The top emitter resistor becomes 0.2 ohms, and the lower becomes 0.34 ohms. Prefered (or nearest available) values are 0.22 ohms and 0.33 ohms.

Rather than disturbing the upper emitter resistor, I placed 2 x 1-ohm resistors, all in parallel with the original 0.33 ohms, which made precisely 0.2 ohms. I left the lower resistor at 0.33 ohms as being close enough.

The voltage appearing across these resistors might alarm those conditioned to believe both halves of the output stage must be identical/equal. Still, hopefully, the above discussion shows the actual, but unmeasurable emitter voltage drops are equal.

All we can do is measure the remaining voltages which come out of this when the meter is placed across them.

So that top and bottom halves average out any remaining error (which is so tiny as to be of no consequence), it is better to measure across emitters. Meters are included in the design for the user to set via fine trimmers (this being necessary due to differences in supply voltages).

0.2 and 0.33 ohms appearing in series requires an 18.9mV voltage drop to set Vq for both halves at 26mV each - this being optimum for minimizing the onset of gm-doubling.

Comparing distortion on the analyser, I thought there must be a fault, as it was considerably better than I expected.
That none should be able to buy or sell without a smartphone and the knowledge in how to use apps
Back to Top
 Post Reply Post Reply Page  <1 9899100101102 345>
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.01
Copyright ©2001-2018 Web Wiz Ltd.

This page was generated in 0.109 seconds.