My attempts at a design of a 3 stage amplifier

The elimination list (continued):
vi) Improper grounding, maybe.
The present grounding of the two amplifier channels, is through mechanical contact of the two large heatsinks that cool the output stages. These large heatsinks are fixed to the bottom of the metal amplifier box with three screws each.

The rectifiers PCB has NO grounding and power supply smoothing capacitors are on the amplifier channels main PCBs.

I wish to thank user CurrentFlow who suggested to me using a Raspberry Pi together with a Pi DAC PRO from IQAudio. After many hours of fiddling, I finally am enjoying music played with this beautiful piece of hardware. I am using Raspbian. Unfortunately, there was a bug involving the detection of the Pi DAC PRO, but now, it is resolved.

I can only say, the audio playback is so resolved, that it seems, as if a band is playing inside my home.

I also discovered there are software packages that turn a Raspberry Pi into a web-based music player. In my case, this is an option, as normally, I prefer a monitor and keyboard attached to the Pi to access it directly.
 
The elimination list (continued):
vi) Improper grounding, maybe.
The present grounding of the two amplifier channels, is through mechanical contact of the two large heatsinks that cool the output stages. These large heatsinks are fixed to the bottom of the metal amplifier box with three screws each.

The rectifiers PCB has NO grounding and power supply smoothing capacitors are on the amplifier channels main PCBs.

vi) improper grounding, mechanical contact is not electrical contact - without this the whole of the amplifier electrics are floating and you are reliant on the either your Rasberry Pi or your DAC to have a signal connection to ground.

If these have +5V supplies via a USB mains plug ground would be represented by the neutral mains connection to an earth stake at a power substation some distance from your house with more resistance in the wiring between that you would have from the earth stake outside your house.

I take it that your chassis and the heatsinks would be connected to this by the earth of your mains plug.

You should measure the voltage between the chassis and the centre tap of your transformer.
 
All right, I measured the AC voltage across the secondaries' centre taps and the amplifier metal box and got 0V in both cases. I remember when I was building the amplifier I made sure this resistance path is 0 Ohms.

The problem I am experiencing is more subtle than it appears. All attempts failed to rectify the hum problem, although the filters for the sensitive amplifier circuitry, brought some improvement.

In an approximate prototype that I built before as a guide, I corrected a monstrous hum issue by connecting an electrolytic capacitor across the two series biasing diodes. If I remember well, the capacitor was a 1000uF one. I think, this amplifier is suffering from the same issue. The VAS's and input bias may contain interference from the mains and some ripple.
 
Looking at one of your simulations, the bias network for the cascode has a 10k resistor feeding about 7 mA from the positive rail into a 12V zener diode D3.

If you add a 100 uF capacitor in parallel you create a low pass filter effect on the positive rail and by-pass the noise generated by the zener.

The simulation does not include any allowance for ripple on the supply rails and you don't want any of this in the collector current of Q5 at the start of the amplification chain.
 
Simulating for the input and VAS bias chain, the attached results were obtained. Clearly, the bias chain adopted by Dougles Self for his Blameless Amplifier has a far superior performance to mine. I am attaching both simulations for comparison.

In the extreme, I can still adopt Douglas Self's two transistor bias together with the centre tapped risistor chain plus 1uF capacitor to ground.
 

Attachments

  • douglas_self_input_vas_bias_filter.png
    douglas_self_input_vas_bias_filter.png
    77.4 KB · Views: 180
  • my_input_vas_bias_filter.png
    my_input_vas_bias_filter.png
    75.4 KB · Views: 180
I suggested the 100uF addition as an addition you could test on the actual hardware to see if there is a subjective reduction in the hum. You have not given an answer on this.

Earlier I had suggested the inclusion of an RC filter in the positive rail which according to simulations would have positive impacts on ripple and distortion by cleaning up the supply to the small signal stages. Again there is has been no answer on whether you have tried this out in your hardware or not.

A reported reduction in ripple on the positive rail to about 500 uV seems to have been forgotten in your latest simulations where this has jumped to 25V - a figure which seems to have been plucked out of the air.

The low pass filtering factor of D3 12V zener with 100 uF in parallel can be increased by making R2 =33k which would reduce the current and noise in the zener.

As far as a copper strap is concerned with patience it could be a diy proposition - see attached.

I have spent as much time as I can afford on this thread so good luck with getting this problem sorted.
 

Attachments

  • Copper Strap.jpg
    Copper Strap.jpg
    146.9 KB · Views: 185
Thanks for the magazine page photo. It clearly illustrates how a copper strap is installed around a toroidal transformer. I tried this procedure using makeshift iron sheets but I got no reduction in hum. I also unbolted the transformer and placed it out of the box without hum reduction.

Had I got a hum reduction using this procedure I would have certainly adopted it.
 
This means I have a small ground loop inside the amplifier box. The input is grounded at the front panel, and the amplifier is grounded at the 0V point between the two large smoothing capacitances. In the same box there is a huge toroidal transformer, which usually leaks a weak magnetic field. This in turn, generates a voltage in any conductive loop it encounters. As a precaution, I fixed the input cabling with tape to make sure the cabling is in contact with the metal chassis. However, I cannot predict from where electrons flow in the metal box, and in this way, a loop can form.

I rerouted the input signal cabling. It seems the hum was injected through parasitic capacitance between the heavy supply tracks and the input cabling.

For now, I am satisfied with the results although there is always room to improve things. I am starting to make the second channel PCBs using the same old fashioned method.

My last post before taking a break from this thread was 416 on page 42 where I suggested following best practice for routing of earth paths. That was back in early March 0/10 for following this up.

Seemingly your signal earth and the supply earth connect to the chassis at separate points in which case your diagnosis of a ground loop was correct.

You can look this up on Wikipedia - had you done this at the time you would have answered your own questions.

If you are using RCA chassis sockets for your inputs, the bodies of these need to be of a type that is insulated from the chassis. The screw-on solder lug for each of these should connect through separate 10 R resistors to a solder lug on the rears of the chassis. Don't earth these to the face plate of the amplifier.

It is possible to screen the transformer using strips of grain oriented steel bent to a circular shape and fitted in overlapping layers around the toroid and kept in place with insulating tape.

The current around a toroid is circular in motion as this follows the direction of the winding - the same applies to the flux field only the shield needs to be in close contact so the induced current travels through the shield in a circle.

There are other aspects of performance that will be improved by screening the transformer.
 
Thanks for the link.

The most sensitive part of the amplifier is the differential pair. The non-inverting input is biased through a 10k resistor connected to ground. The inverting input is not directly biased through a resistor to ground, but current has to pass through a large electrolytic capacitor.

Studying the PCB ground track feeding both inputs, one observes that ground current flows along a partial loop. Such a loop interacting with a varying magnetic field, certainly generates an electrical voltage along it. The latter, is between the non-inverting and inverting inputs, which means, such a voltage, is amplified by the amplifier's open loop gain, which is very high at 50Hz to 100Hz.

Therefore, the ground track has to be shortened and the signal ground lifted off power ground with a 10 Ohm resistor.

I will consult other amplifier circuits to study their input grounding and filters to modify this amplifier's input. I am certain the elimination list is getting shorter, which means, finally this problem will be sorted out.

Post Scriptum:
I shielded the toroidal transformer with a stack of about 22 wire loops but the hum remained. I did this to avoid having to buy a copper strap to discover it is ineffective afterwards. The straps I found online are quite expensive. I also disconnected the inputs from the chassis's input socket board. Again, the hum remained. Furthermore, I unbolted the toroidal transformer and placed it as far as possible from the circuit boards, and still, the hum remained.

The latter, is between the non-inverting and inverting inputs, which means, such a voltage, is amplified by the amplifier's open loop gain, which is very high at 50Hz to 100Hz.
This conclusion is incorrect, since such a generated voltage, has to appear unattenuated across the non-inverting and inverting inputs , which is impossible due to the existence of the feedback network.

If this hum proves to be impossible to eradicate, it will be a real waste of components and time for me, and whoever contributed to this thread.

What is strange in my PCB for the input stage and VAS, is the fact that it didn't strike my head that I was effectively inserting a loop between the two inputs. The latter should have been ground referenced to exactly the same point on the ground track. The so formed loop is about 3cm away from the transformer, implying, a strong exposure to leakage magnetic flux.
 
You need to avoid connecting the signal earth directly to the supply earth.

The idea of the T junction in the attachment is to allow the a.c. charging currents for the capacitors to neutralise one another on the arms of the cruciform - away from the centre tap of the transformer which is the next dirtiest connection to earth.

You should follow the precedence in the attachment I posted for speaker earth returns for local decoupling capacitors, and small signals.

The cleanest earth has to be last in the chain. The group including the non inverting input earth, the inverting input earth and the biasing earth for the input stage.

If you get this wrong for instance the supply capacitor earth is at the end of the chain then the resistance between these points will drop a greater voltage than any other earth return will do.

You are looking at minute voltages but these will be accepted as signal voltages which will be sent to the inverting input of your amplifier. In terms of scale comparing the signal currents the supply capacitor earth is the flow of a landslide compared with small signal currents.

The effects of incorrect earthing could easily be masking noise from the radiated field of the toroid in which case your substitute test with wire proves nothing.
 
Administrator
Joined 2007
Paid Member
It might be worth you going back to the link in post #361 and having a look at the linked thread:

My attempts at a design of a 3 stage amplifier

When it comes to grounding, even connecting two wires to the same point isn't good enough if that point has ripple current flowing through it... that current will be enough to generate a voltage actually within the point you have the wires connected. There will be a voltage gradient across and within the connection.
 
I am attaching the PCB layout for the input stage and VAS. The ground loop I mentioned earlier in this thread is shown in green. This has a radius of around 3cm and the PCB is less than 4cm away from the transformer. The non-inverting input is on the left hand side. I am thinking about breaking the base track of the non-inverting input to use a thin wire, and route this wire along the right hand side and top of the green track until it reaches the input filter. The other end would be soldered to top end of the broken track. This way, I am hoping to get two conductor segments with equal and opposite voltages induced by the leaking magnetic flux. If I also break the green ground track at the top left hand side just before the input filter, I can use a thin twisted pair of wires and take ground contact from the ground connection of the large feedback electrolytic capacitor. The other end would be soldered to the grounded beginning of the input filter. Instead of a thin twisted pair I can also use a thin coaxial cable like those used for VCRs.

Together with the above, I will separate the signal ground and power ground with a 10 Ohm resistor as recommended.

If these mitigations prove insufficient, I will re-examine the points from where I can take power from the power supply section.
 

Attachments

  • input-vas-driver-pcb.png
    input-vas-driver-pcb.png
    12.4 KB · Views: 158
Last edited:
You could try rearranging your nfb divider arrangement to fit with your simulation file.

You could try rearranging your nfb divider arrangement to fit with your simulation file.

I should have said arrange your nfb divider along conventional lines. Forget about that for the time being.

Anyway looking at the image you posted I did not see where is the RC decoupling of the supply rails shown in the simulation attached to post 482.

Without these you will have some ripple passing through the bias resistors for your input stage which is not helped by using values suitable for amplifiers running off low voltages supply rails and with 80V rails passing several m.A. where the need is measurable in u.A.

A couple of resistors in your image are physically large to bridge wider spans on your pcb. It seems these could be wire-wound types - if so change them.
 
Anyway looking at the image you posted I did not see where is the RC decoupling of the supply rails shown in the simulation attached to post 482.

Without these you will have some ripple passing through the bias resistors for your input stage which is not helped by using values suitable for amplifiers running off low voltages supply rails and with 80V rails passing several m.A. where the need is measurable in u.A.
The RC rail decouplers are on another small PCB that is bolted to the PCB I attached.

A couple of resistors in your image are physically large to bridge wider spans on your pcb. It seems these could be wire-wound types - if so change them
The physically large capacitors are 3W metal film types.

Post Scriptum:
Estimating the leakage magnetic flux I found it is about 1.5mT. This calculation was based on the assumptions that leakage flux is active over a circle of radius 5cm, and that, a voltage of 1mV peak is induced.

For a transformer the peak magnetic intensity is about 1T. This means, a leakage of 0.15%.

The 1.5mT leakage magnetic flux density is an overestimate. This would produce a 50mV peak mains hum. A 10mV ripple is more realistic in my case, which means, the 1.5mT figure becomes 1.5e-3/(50e-3/10e-3) = 0.3mT.

High resolution, very slow, LTSpice simulations are confirming, that the ground link linking the non-inverting and inverting inputs, is very susceptible to any voltages induced there. The link joining the current mirror output to the VAS, is on the other hand, not vulnerable, as expected. This confirmation by LTSpice is motivating me to first make changes to the ground link between the inputs, so that, any voltage induced is cancelled out.

Simulating for an improper connection between the power supply smoothing capacitors, did not yield much information about issues with the output. This does not mean, however, that this can be abused.

The other change to the circuit will be a 10R ground breaker between the power supply and sensitive input parts.
 
The ground current path from your cascode zener diode ZD1 reference connects through the 0 V for the inverting and non-inverting input. This should be vice-versa.

I note this network ground is close to the nfb decoupling capacitor so connected the resistance between these two points in comparison to that from the network to signal ground has to be considered 'susceptible' in terms of micro voltage drop seen as a voltage signal.

I have commented previously about the use of zener diode biasing. It should have a capacitor in parallel to pass the noise this generates to ground.

Better still would be to bias the cascode via a resistor divider with a parallel capacitor to augment the filtering of the RC in the supply rail.
 
Last edited:
This has a radius of around 3cm and the PCB is less than 4cm away from the transformer
I'd say ideally you'd have 6" or more between transformer and any low-level signal circuitry. Magnetic fields fall off as the cube of distance(*), so the difference between 3" and 6" separation can be substantial.

(*) strictly speaking only in free space, and only when the magnetic component is smaller than the separation distances involved.
 
The closed loop gain of the power circuit is close to 69 - a point which I made earlier.

In the technical manual for the original circuit the power amplifier is fed by an op.amp having a closed loop gain of two.

If that is still in circuit the total amplification is 138 times. There is no volume control to allow a short of the input terminal shorted to earth to test the noise.

If the amplication is 3-6 times greater than usual and the transformer also has a power rating three times greater then you are upping the ante.

Actually the closed loop gain of the original Wharfedale circuit is 27 versus 69 which is roughly 2.5 times greater rather than 3 - nonetheless still over the top.
 
The simulated gain with a 1000mV peak input signal is 47.9. This is Vout/Vin, I am not using dBs.

The gain of the original amplifier circuit is 33dB. Converting this using an online dB to raw gain converter, the value is 44.67. This is very close to what I am simulating.

I have never considered the possibility that the input differential pair may be causing the hum issue because they are not matched. LTSpice seems not to simulate very scrupolously the effect of Vce on dynamic gain (beta). This means, if there is some hum component in the tail's current source, the differential pair may not be completely cancelling it.

I am using two BC337-40 transistors for a differential pair, maybe, those are not up to the task.

The original design used an opamp, but in the beginning of this thread, I was strongly suggested not to go that way. A good quality opamp, usually has good power supply rejection, high gain and low distortion. All of these could have been used to simplify this amplifier circuit. Such a design would have only needed two high voltage transistors as used currently in the VAS stage. I found an amplifier that uses such a technique with a very low distortion figure comparible to Douglas Self's famous Blameless amplifier.
 
Last edited: