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 Club

 

Open to all owners plus those contemplating the purchase of a Graham Slee audio product wishing to use our loaner program: join here
Subscribe to our newsletter here (Rules on posting can be found here)


1970s Design Indulgence

 Post Reply Post Reply Page  <1 99100101
Author
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 9269
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 08 May 2020 at 8:01pm
References:

Distortion in Complementary-Pair Class-B Amplifiers, B M Oliver, Feb 1971
Audio Power Amplifier Design, D Self, 6th ed, 2013
Designing Audio Power Amplifiers, B Cordell, 2011
Prevent emitter-follower oscillation, M Chessman & N Sokel, Electronic Design, June 1976
Motorola and On-Semiconductor relevant transistor datasheets
https://www.diyaudio.com/forums/solid-state/101745-bob-cordell-interview-bjt-vs-mosfet-300.html#post2052523
Not simple enough for Google-Bot to understand...
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 9269
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 11 May 2020 at 4:21pm
One now glaringly obvious observation is how the HF falls away after a knee at a faster rate before crossover. It is very similar to that seen with uncompensated op-amps, such as the NE5534 and OP37, which are only stable at gains greater than unity, and one technique is to increase noise gain just before the droop.

This amplifier needs more high-frequency open-loop gain, so that, with negative feedback applied, the high-frequency distortion is reduced.

Compensation capacitance might go lower but with a phase margin penalty. Even so, it only gives 18dB 20kHz loop-gain. The local negative feedback resistor has to go up in value to let that happen. But because of that, the open-loop frequency response then falls to about 8kHz, and it is advantageous to extend it as far up the audio spectrum as possible.

Taking the compensation cap down to say 47pF, then places us firmly in oscillation territory. But if we trick the noise gain by placing a small capacitor across the negative feedback divider resistor (the one which goes to ground), the noise gain will "lever" the curve upwards, countering its rate of fall, and making the phase margin just shy of 60 degrees.

The loop gain at 20kHz becomes 22dB, so we get 4dB more NFB and double the open-loop frequency response.

I'm hoping this improves pronunciation of esses whose harmonics, which are either side of 20kHz, don't get emphasised as much by harmonic distortion.
Not simple enough for Google-Bot to understand...
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 9269
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 12 May 2020 at 6:51am
In showing you before and after pictures, conditions should be identical. Here, the "before" image was taken at 1dB lower output than the "after" ones. You will have to trust me that the "before" results at -21dB would not be much different from the -22dB shown here.

The power output into 8 ohms corresponding to -21dB here, is 1/3rd of a watt (333mW), which might not appear to be much. Still, a lot of valuable information/musical-detail emerges into the listening room at that power, during regular listening.

It is also deep into the crossover region where some of the worst distortions in solid-state make themselves known. Enthusiasts are often preoccupied with what happens at clipping and comparing it with valve amplifiers. Still, if you are achieving those levels, your neighbours probably hate you.

The meters, whose primary purpose is to set and monitor Vq, also spill the beans on normal listening power. True, they average and don't catch the transients, but they rarely reach above the 40 mark, which is 1W, while enjoying some rock in the average listening room.

The first two images show before and after the new output stage and NFB mods.


Before



After

The second harmonic and higher harmonics are much reduced. The third harmonic is unavoidable because a push-pull exhibits a cubic function of non-linearity.

This time, I have included results for 10kHz and 20kHz. The 10kHz performance isn't far from what it did at 1kHz before.



The 20kHz results look quite decent. It is where a rapid increase in distortion is expected due to the reduction in the amount of negative feedback. But by allowing a higher cut-off, bought by a narrower phase margin, the additional negative feedback it affords led to a respectable harmonic distribution.


Not simple enough for Google-Bot to understand...
Back to Top
BAK View Drop Down
Senior Member
Senior Member
Avatar

Joined: 14 Mar 2010
Location: Kentucky, USA
Status: Offline
Points: 1047
Post Options Post Options   Thanks (0) Thanks(0)   Quote BAK Quote  Post ReplyReply Direct Link To This Post Posted: 12 May 2020 at 1:30pm
Looks to be a good improvement over the AP 1/3w 1kHz distortion product ratios shown on page 99.
Bruce
AT-14SA, Pickering XV-15/625, Technics SL-1600MK2, Reflex M, Lautus, Technics SH-8066, Dynaco ST120a, Eminence Beta 8A in custom cabs;; Using Majestic DAC
Enjoy Life Your Way!
Back to Top
Graham Slee View Drop Down
Admin Group
Admin Group
Avatar
Telling it as it is

Joined: 11 Jan 2008
Location: South Yorkshire
Status: Offline
Points: 9269
Post Options Post Options   Thanks (0) Thanks(0)   Quote Graham Slee Quote  Post ReplyReply Direct Link To This Post Posted: 31 May 2020 at 2:31pm
Power supply smoothers, being electrolytic aluminium cans, are limited in their high-frequency performance.

The above should be evident by their ripple current rating, but it is often ignored because educators seldom mention spikes.

Morgan Jones differs, most probably because he digs deeper into what a power supply does, and offers mathematical and graphical proof.

The electrolytic is called upon to do 10 times more current than the load requires, which suggests that most of us designers should hang our heads in shame.

Ripple voltage and current

The drawing reproduced here from his book "Valve Amplifiers" enlightens us that ripple current flows for a fraction of time each half cycle, and might well be for only one-tenth of the 100Hz rectified cycle. To comply with the law of charges, Q = CV = IT, the current must be 10 times the load current.

My use of the Philips/Vishay BC Components electrolytics as smoothers is justified by the datasheet, which indicates their characteristics can accommodate, higher than rated, ripple current. Maybe that wasn't so for the other types I used.

The illustration also graphically shows that with reduced amplifier current demand, the current spikes become shorter: as V ripple becomes shallower, t becomes narrower.

Luckily, the ripple current falls with reduced amplifier current demand, but this shows that the power supply smoothers (reservoirs) always have a tough time, not just at full power.

At high power, the ripple current time can equate to a high energy 1kHz pulse, and at low power demands, a lower energy but higher equivalent frequency pulse.

Class B operation (or class aB as here), must reflect its current-frequency energy to the power supply, or as Kelvin noted, flows in circles.

Just like in the PSU1 Enigma, we must chart the path from power supply "origin," around the circuit, and back to the "origin."

Without sufficient decoupling (local or primary power supply energy storage), the ripple voltage and ripple current would make its impression on the sound, and you would hear a buzz.

Local decoupling of the input stage, and constant current, via the bootstrap arrangement, to the voltage amp stage, significantly reduces buzz. Placing an ear close to the speaker confirms this, and so do the noise measurements.

However, if we make a change to the power supply circuit, a quality difference can be heard.

We cannot escape the basics as described above without voltage regulation. A suitable voltage regulation circuit for this amplifier would at least double its complexity. Switched-mode technology would also supply the solution, being inherently stabilised.

Including either would go against the spirit of this design, so in perfecting the power supply, we must look elsewhere.

Could there be anything else we might have overlooked? One answer might be back EMF in the transformer caused by rectifier diode switching, and this issue has been raised earlier in this topic.

The stock solution is to bypass each diode (or bridge rectifier section) with a 10 - 100 nF capacitor.

Another solution is to place a 10n capacitor across the bridge rectifier AC input.

What these do is slow down the spike produced below radio frequencies. The spike still exists, but at a frequency which is non-transmissive.

Transformer back EMF spikes

The problem is that the problem is still there. Although the transformer is supplying the current, the workings are just the same as switching off a relay, in that a reverse EMF takes place.

This reverse EMF takes place every half cycle, and the spike it produces not only radiates into space, but radiates by the more natural path, in the wiring, and must, therefore, exist at the terminals of the smoothers.

If we don't remove the spike, it doesn't matter at what fundamental frequency it has, it is still a spike, and a spike which adds to the spikes already there due to ripple current.

If an electrical current cannot return to its origin, does it flow in the first place? I think Faraday might have made such an observation. Lightning can only happen when the right conditions exist.

If the amplifier is called upon to deliver a facsimile of the input signal to the load, but at the same time, a series of power origin spikes occur, do these mix, and produce a signal to the load which differs from the input signal?

Where exactly are these spikes? Is it conceivable that they change their positions relative to time and their amplitude, depending on the signal the amplifier is "trying" to transfer to the load?

Should we therefore try and ensure no such spikes exist?

Another article I once read by Morgan Jones referred to damping the spikes produced by bridge rectifier diodes, in which a rather "weedy" snubber was shown to be far more effective. If I remember correctly, he was subjected to heaped criticism for it.

I constructed a rather crude simulation model of the power supply circuit to show the back EMF spike from the transformer when stimulated by an arbitrary signal source. The results are shown in the image.

The value of inductance might not be the same in reality, because measured inductance of a transformer secondary can differ considerably by what is on its primary side. At some point, the primary can be effectively shorted - when the voltage is at its peak - and effectively open circuit, when the voltage is at its minimum.

I chose the open circuit inductance because that is close to where the diodes should be switching off: this measured 80mH.

The rest of the circuit is as it exists, and the load can be varied but makes no difference.

With no snubber, a spike can be seen near 80kHz; this incidentally is in the frequency range where EF output stages can suffer parasitic oscillation.

With the conventional 10nF snubber, the spike falls right in the vocal harmonics region, or the lower "esses."

Either spike indicates 10 - 30 times the amplitude, and some portion must be transferred to the smoothers once the diodes start to conduct again (or otherwise transmit via circuit capacitance).

I use the term "wear-down" to describe the smoothers getting "tired" of handling such spikes, in the same way that the ripple current can be ten times the load current.

What I observe is that after a rested state, the sound gradually deteriorates after the power is reapplied.

Perhaps if the spikes can be attenuated or smoothed, the smoothers might be able to handle the resulting lesser current demand?

The simulation revealed that a "weedy" snubber of 39k in series with 220 pF, reduced the spike to a curve containing less energy. It also places it lower in the parasitic oscillation region and places it above the audible range.

The frequency calculates from the transformer inductance (at the moment in time) and the capacitance value of 220 pF. The 39k resistor decides the amplitude.

The power consumed in the resistor depends on the phase angle, and I'm not even going to venture there. SPICE does the maths. I conjecture that at 50 volts RMS transformer output, it dissipates 0.06 watts, which might act as a brake.

So, the best thing is to give it a try, and give it some time, and report on my findings (yet again).

Edited by Graham Slee - 31 May 2020 at 2:32pm
Not simple enough for Google-Bot to understand...
Back to Top
 Post Reply Post Reply Page  <1 99100101
  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 0.063 seconds.