Hello Rob, s?ndag 29 mars 2020 skrev Rob Jarratt <robert.jarratt at ntlworld.com>: Do you know the nominal output rating for the supply? It might be the case that even with load board the load is uneven. But it sounds less likely. Can be hard. But a good idea is try to minimise the length of wires. Maybe use designators for networks instead of wires to make the schematic more readable. On the other hand I was reading the post on the iPad. bench supply will end up at line potential. With the protection transformer in place the circuit will be left floating in relation to earth. With a variac you can then vary the input and keep it within safe limits. If you don?t have a protection transformer and variac then another bench DC supply that can give up to 100 V can be used instead. Absolutely right. I should have been more explicit about where to introduce external supply. But as you say, lift R32 to isolate the switching controller and feed in the bench supply current at VCC. Check the data sheet for what is approriate voltage. Good luck! /Mattis

Thanks Matt, I just found that bit in the documentation and realised I am not generating enough load. I will have to add more. I did connect the 5V load because as you say it wasn't connected. I can't test it now until I get another fuse, but I am sure I didn't see this happen, although it seems to be quite hard to tell.

I found time to follow Mattis?s suggestion today and I got some interesting results. I powered the UC3842 with about 16V from a bench power supply. I lifted R32 so that the transformer would not supply it. I then used an isolating transformer to power a variac and applied the variac to the AC inlet. I also used a load board from a MicroVAX 2000 and an old RD53 disk as the load, so there should be enough load. I found that I can vary the AC input up to a maximum of about 40VAC before the SCR triggers, the 5V output reaches about 400mV. If I raise the AC input more slowly, it will usually cut out before that, around 30VAC. I noticed that the inrush thermistors also get quite hot at these low AC voltages, I don?t know if this is because of the relatively low AC supply voltage, or if this indicates a problem of some kind. The voltage coming out of L3 into the T1 ?bounces? somewhat. I guess this is because the AC input is only 20V or so, or it may be expected ripple from the smoothing capacitors? In the description below, the peaks of the bounces are used. Throughout the variation from 0VAC to 40VAC the duty cycle of the oscillation of the UC3842 output does not change, I guess because the output voltage has not reached its target value. With the AC input at about 25VAC the circuit seems to be stable (apart from the bounces mentioned above). At this supply voltage, the voltage at the source of Q1 reaches 2V. The current sense resistor is 1 Ohm, which means 2A must be flowing through it at that time. When the Q1 source is at 2V, the other end of R14 is at about 0.5V, which is just below the trigger voltage for the SCR. This makes sense because R14 and R15 form a voltage divider that looks to be nominally 25% of the Q1 source. Given the SCR nominally triggers at about 0.8V, this means that the current sense resistor is set to trigger the SCR at about 2.5A, I think. This would suggest that the duty cycle on Q1 is too high and causing too much current to be drawn. So presumably the feedback to the UC3842 is not working correctly. I tried setting the AC input at 120V and using a one-shot sample. Q1 is switched for about 30ms and then there is a spike on the SCR gate to 2V and it triggers. The gate voltage then remains at 1V. However, there is no spike across the current sense resistor (R13), so I don?t know if the spike is because the SCR is being turned for some other reason. There is nothing unusual on the anode of D19 to cause it to trigger due to avalanche breakdown. I got the same result when the AC input was 220V. I wonder if the SCR is behaving slightly differently because I have lifted R32? Since there might be a feedback problem, I looked at the VFB input to the UC3842 when doing a one-shot test at 240VAC. I can see VFB steadily rise over the period when Q1 switched, up to a maximum of 4V. I don?t really know if this is how it should behave though, but it seems to make logical sense. During all that time the duty cycle of Q1 does not change. I am not too sure where to go from here. I hope the above makes sense. I would appreciate any further thoughts. Regards Rob From: Rob Jarratt <robert.jarratt at ntlworld.com> Sent: 29 March 2020 11:40 To: 'Mattis Lind' <mattislind at gmail.com>; rob at jarratt.me.uk Cc: 'General Discussion: On-Topic and Off-Topic Posts' <cctalk at classiccmp.org> Subject: RE: VAXmate PSU Thanks, I do have an isolating transformer as well, so I will use that. I have done previous tests with a bench PSU on the UC3842 and found it needs 16V to get going. I just checked the Technical Description and it says the minimum current on the 5V output is 6.4A and on 12V it is 0.17A. I checked my load board and it is only going to sink 3A, so I need more load as I think the IDE disk is not going to be much. Would insufficient load really cause it to shutdown so quickly though? Not sure I understand your comment about ?designators for networks?, is there an example you can point me at? One of the things I have tried to do, but clearly not very successfully, is to minimise the lengths of the wires. Regards Rob From: Mattis Lind <mattislind at gmail.com <mailto:mattislind at gmail.com> > Sent: 29 March 2020 08:51 To: rob at jarratt.me.uk <mailto:rob at jarratt.me.uk> Cc: General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> > Subject: Re: VAXmate PSU Hello Rob, s?ndag 29 mars 2020 skrev Rob Jarratt <robert.jarratt at ntlworld.com <mailto:robert.jarratt at ntlworld.com> >: Thanks Mattis. I do test it with a load, although to be honest I forgot to do this when I took the measurements yesterday. I use a modern-ish IDE disk and a load board from a MicroVAX 2000 as the dummy load. I don?t know if that is sufficient. Do you know the nominal output rating for the supply? It might be the case that even with load board the load is uneven. But it sounds less likely. I don?t know enough about PSUs to make the secondary side drawing more logical unfortunately. Can be hard. But a good idea is try to minimise the length of wires. Maybe use designators for networks instead of wires to make the schematic more readable. On the other hand I was reading the post on the iPad. I have a variac and a bench power supply, so I could do what you suggest. To stay safe you need a protection transformer as well. Otherwise the the bench supply will end up at line potential. With the protection transformer in place the circuit will be left floating in relation to earth. With a variac you can then vary the input and keep it within safe limits. If you don?t have a protection transformer and variac then another bench DC supply that can give up to 100 V can be used instead. Could you be a bit more specific about where to apply what, so I don?t do it wrong or damage something? Would you put the bench PSU across the UC3842 Vcc and Gnd pins? I am not sure what would happen if the normal supply to the UC3842 was still in place with the bench power supply also trying to supply power. Would it be wise to lift R32 so nothing conflicts with the bench power supply? Absolutely right. I should have been more explicit about where to introduce external supply. But as you say, lift R32 to isolate the switching controller and feed in the bench supply current at VCC. Check the data sheet for what is approriate voltage. Good luck! /Mattis Thanks Rob From: Mattis Lind < <mailto:mattislind at gmail.com> mattislind at gmail.com> Sent: 29 March 2020 06:39 To: <mailto:rob at jarratt.me.uk> rob at jarratt.me.uk; Rob Jarratt < <mailto:robert.jarratt at ntlworld.com> robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts < <mailto:cctalk at classiccmp.org> cctalk at classiccmp.org> Subject: Re: VAXmate PSU Hello Rob, l?rdag 28 mars 2020 skrev Rob Jarratt via cctalk <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> >: I have posted here a couple of times because I have a failed VAXmate PSU. I have just posted a little bit more information here: https://robs-old-computers.com/2020/03/28/further-analysis-of-the-vaxmate-h7 <https://robs-old-computers.com/2020/03/28/further-analysis-of-the-vaxmate-h7270-psu-failure/> 270-psu-failure/ with some scope traces and a greatly improved schematic. Although the schematic is likely to have errors still. Unfortunately, a stray scope probe ground lead blew the fuse so now I have to wait for a new fuse to arrive before I can continue work. I would really like to know if all the spiking I am seeing is to be expected, and any suggestions why it appears to be detecting an overcurrent? There do not appear to be any shorts on the secondary side, but that could be wrong of course. I don't know if a genuine short anywhere would cause it to trip the SCR quite so quickly (within 20ms of the switching transistor starting to switch). This is fly back design and I would expect some spiking when the transistor shuts off. Then for over-current. It might be so that there are over-voltage protection on the outputs that kicks in. A crowbar that short circuits the output. It looks like there is such a circuit on 5 and 12 V. But to be honest the output circuit schematic is hard to read. If you have no load or little load or un-even load the PSU might hae problems to regulate. I know for fact that the PSU in the MicroVAX 2000 need to have a dummy load when no hard drive is installed otherwise there will be uneven load which it has hard time to handle sonce the output regulation is based on the sum of the outputs somehow. It will trip the crowbar on over voltage on one of the outputs otherwise. What if you supply the control circuitry on the primary side using a bench lab supply and then connect a protection transformer and a variac in series to the normal AC inlet. Slowly increase input AC voltage while monitoring source voltage and output voltages. At what AC input voltages does it trip? What is the output voltages at this point? If both voltages exceed normal and the crowbar trips I would think that the feedback network somehow reports to low output voltage to the control circuitry. Maybe the opto coupler is bad? Sorry. A lot of guessing here. But it is hard to tell withour more measurements. /Mattis Any thoughts gratefully received. Thanks Rob

via On-Topic power load. will transistor(s)). lead to or energy resources I have to say that when I was thinking about this, I did wonder if the problem was that it was trying to raise the output voltage with not a lot of input, and that therefore the duty cycle would be too high. I will remove the variac from the equation. For the record I was seeing a duty cycle of about 50%. In later testing at 240VAC the duty cycle does seem a lot lower. rated output. various the towards a or The reference. I have already done a ringing test on the main transformer, and I think that it is OK. One of the windings does not ring very well, but I think it is one that has few windings and supplies the on-going power to the primary side once the PSU has started up. I have also done a bit of testing on the secondary rectifiers, but not found anything so far. I will look at secondary side again more closely. I have obtained a scope trace as you suggest. R32 is still lifted so the UC3842 is powered by the bench PSU, but I am using the full 240VAC (no variac). The channels are: 1. Ch1. 555 timer. 2. Ch2. D19 Anode 3. Ch3. D19 Gate. 4. Ch4. Q1 Source. The picture is here: https://rjarratt.files.wordpress.com/2020/04/h7270-control-pulse-width-modul ator.png the awfully the You are correct, I have mixed up the values with R30 and R31, the correct value is 15K. I have updated the schematic, and rearranged it to look diagrammatically more like a sample diagram in the TI datasheet for the UC3842. The updated schematic is here: https://rjarratt.files.wordpress.com/2020/04/h7270-control-pulse-width-modul ator.png Thank you, I am sure it will be simple when I find the problem, and I am learning a lot.

haven't Sorry, that looks like a cut and paste error, here is the link to the scope picture https://rjarratt.files.wordpress.com/2020/04/h7270-primary-scr-trigger.png I used a 100ms timebase for the capture and then "zoomed in" a bit your display charged can traces at a just

via <cctalk at classiccmp.org> supply, a something +V action. the but crowbar - Oh dear! After Brent's question about D19 anode, I realise that the probe was connected to the cathode! I have now done it again with the probe connected to the anode. I have taken two images of the same capture, one at low resolution to show the overall behaviour https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-1.png And one zoomed in to show what happens when the SCR shuts down. https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2. png The channels are the same as before, namely: Ch1. 555 timer. Ch2. D19 Anode (now corrected as it was previously the cathode!) Ch3. D19 Gate. Ch4. Q1 Source. I got an earlier trace which showed the D19 anode at 9V, which is under the Undervoltage Lockout threshold, but I have not been able to repeat it. I don't fully understand the debate about using the variac. However, my measurements appear to suggest that when I use the variac the SCR triggers because of what appears to be a genuine overcurrent detected by R13. I think this is because the duty cycle at low AC input voltages is 50% (rather than about 10% or less as per the trace I have just taken), and I measured 2V across R13, which does seem to be enough to trigger the SCR. When I use 220VAC, the voltage across R13 does rise to 6V, which should also trigger the SCR I think, except that the peak last a lot less and so perhaps the fact that the 6V last for a brief period is insufficient to trigger it? I have seen the suggestions to study the waveforms at a much higher resolution. What I am doing is setting the overall timebase in the 100ms range so that I can trigger on when the 555 starts to oscillate and capture the whole period of operation until the SCR triggers. I can then zoom in, as can be seen from the trace provided in this email. I hope that is good enough, or am I missing some problem with doing it this way? I would like to follow Mattis's suggestions (and other people have said it too) to break the feedback loop, but it does look difficult to know how best to do it. I also understand Brent's suggestion that the gate spike is just the result of the SCR triggering, rather then the cause of the trigger. I had wondered if that might be the case. Regards Rob

I did have this in mind but it seemed to me that the sampling was fine enough. To be sure though, I just now changed my trigger to capture the drop in the D19 anode voltage and set the timebase to 10us, but the results looked identical. I couldn't see any spikes that were different to the pictures I posted last night. Thanks Rob

I will look at all the suggestions, particularly of a failure on the secondary side. Something must have burned up, because there was a distinct burning smell after the initial failure, although I have never been able to see any physical damage to anything, despite looking many times. But the thing that really puzzles me is that, after correcting the probes to include the D19 anode, there doesn?t seem to be anything that would cause D19 to trigger. Am I reading the trace wrong? Thanks Rob From: Mattis Lind <mattislind at gmail.com> Sent: 08 April 2020 07:42 To: rob at jarratt.me.uk; Rob Jarratt <robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org> Subject: Re: VAXmate PSU Den ons 8 apr. 2020 kl 00:34 skrev Rob Jarratt via cctalk <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> >: via <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> > supply, a something +V action. the but crowbar - Oh dear! After Brent's question about D19 anode, I realise that the probe was connected to the cathode! I have now done it again with the probe connected to the anode. I have taken two images of the same capture, one at low resolution to show the overall behaviour https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-1.png And one zoomed in to show what happens when the SCR shuts down. https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2. <https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2.png> png The channels are the same as before, namely: Ch1. 555 timer. Ch2. D19 Anode (now corrected as it was previously the cathode!) Ch3. D19 Gate. Ch4. Q1 Source. I got an earlier trace which showed the D19 anode at 9V, which is under the Undervoltage Lockout threshold, but I have not been able to repeat it. I don't fully understand the debate about using the variac. I am not going to debate this either since I know what I have been doing for years and it works perfectly well for me. I have fixed the bigger PSUs in a VAX 11/750 (one broken switch transistor and multiple broken output rectifiers). PSU in NORD-10/S (most carbon composition resistors had gone out of spec). PSUs in many smaller machines as well. I prefer to work in circuits where I can fiddle around without the danger of getting killed all the time. Regardless of use of HV differential probe it can be dangerous. Running it on 50VAC with a protection transformer do expose a lot of problems already and you can poke around safely in the PSU. I have not yet seen a problem that wasn't seen at low voltage, but I expect there could be semiconductors that experience breakdown that occur at lower than specified voltage. However, my measurements appear to suggest that when I use the variac the SCR triggers because of what appears to be a genuine overcurrent detected by R13. I think this is because the duty cycle at low AC input voltages is 50% (rather than about 10% or less as per the trace I have just taken), and I measured 2V across R13, which does seem to be enough to trigger the SCR. When I use 220VAC, the voltage across R13 does rise to 6V, which should also trigger the SCR I think, except that the peak last a lot less and so perhaps the fact that the 6V last for a brief period is insufficient to trigger it? On the issue of duty cycle. If we look at this from the start up perspective rather than the steady state perspective. At startup there are no stored energy in the output filter capacitors. The voltage on the output is thus 0. As soon as the PSU is doing its first switching pulse energy is transfered as the main switch transistor is cutting off. The energy is transfered into the capacitor and into the load. The voltage is starting to increase. The duty cycle generated by the PWM circuitry is in pure relation to the voltage error, i.e. the difference between output voltage and reference voltage. In essence it is a P-regulator. When there are 0 Volt out the duty cycle will be at the maximum. Nothing strange about that. But what is maximum duty cycle? It depends on the circuitry used. The UC3842 can do up to almost 100% duty cycle. However it may be wise to limit duty cycle in a flyback design so that the transformer is not saturated. I am not sure if there is some kind of duty cycle limitation in this circuit though. So if it can handle 50% duty cycle at startup it should be able to handle it at any time. Besides it would be incredible weird to design a circuit to use a 10% duty cycle at its standard operating point and detecting over current at 50%. Then you have much less head room for load and input variations. I am more or less convinced that what you see on the primary side is a result of some kind of fault on the secondary side. A very common problem is short-circuit rectifier diodes on the secondary side (D12, D11, D21, D22, D23, D24). They can be difficult to measure correctly in circuit since the resistance of the secondaries of the transformer is so low. Depending on type you can either desolder them completely or just lift one end of them. My experience is that electrolytic capacitors seldom short circuit. They probably boil and explode instead. Tantalum capacitors often short circuit. Some of them goes into fire other just stay short circuit. So check for tantalum capacitors and try to measure them for short circuit. You have a crowbar on the secondary side. Are you sure that one hasn't triggered? If you still run on variac you can disable the crowbar circuit by removing the SCR and ramp up the voltage slowly to see if that makes any difference. Breaking the feed back loop: R23 seems to be in the feedback path. If you lift it and insert a voltage from a lab supply here you could simulate the output voltage and study the behaviour of the UC3842 for different feed back voltages. You will see that it will stay on max duty cycle up until close to the nominal voltage and over a very small span change to almost no pulse out at all. This is due to the gain of the circuit. /Mattis I have seen the suggestions to study the waveforms at a much higher resolution. What I am doing is setting the overall timebase in the 100ms range so that I can trigger on when the 555 starts to oscillate and capture the whole period of operation until the SCR triggers. I can then zoom in, as can be seen from the trace provided in this email. I hope that is good enough, or am I missing some problem with doing it this way? I would like to follow Mattis's suggestions (and other people have said it too) to break the feedback loop, but it does look difficult to know how best to do it. I also understand Brent's suggestion that the gate spike is just the result of the SCR triggering, rather then the cause of the trigger. I had wondered if that might be the case. Regards Rob

I did use a faster timebase, 10us, triggering on the negative edge of channel 2, and it looked identical to this: https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2… Admittedly the trace you see was taken on the slower timebase and then zoomed in, but there was no difference, the result was identical. Regards Rob From: Mattis Lind <mattislind at gmail.com> Sent: 09 April 2020 05:52 To: rob at jarratt.me.uk; General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org>; Rob Jarratt <robert.jarratt at ntlworld.com> Subject: VAXmate PSU onsdag 8 april 2020 skrev Rob Jarratt <robert.jarratt at ntlworld.com <mailto:robert.jarratt at ntlworld.com> >: I will look at all the suggestions, particularly of a failure on the secondary side. Something must have burned up, because there was a distinct burning smell after the initial failure, although I have never been able to see any physical damage to anything, despite looking many times. Aha. Don?t think I seen you writing about that before, or did you? It might be very hard to find the source some times. Even just a small burn will give quite some smell. Check ALL semiconductors very carefully. But the thing that really puzzles me is that, after correcting the probes to include the D19 anode, there doesn?t seem to be anything that would cause D19 to trigger. Am I reading the trace wrong? It is very hard to tell from the traces what is going on since the resolution is too low. Use a faster timebase. 5 or 10 microseconds. Find out if you can trigger on something that happen only when it stops. Like channel 2 negative slope. /Mattis Thanks Rob From: Mattis Lind < <mailto:mattislind at gmail.com> mattislind at gmail.com> Sent: 08 April 2020 07:42 To: <mailto:rob at jarratt.me.uk> rob at jarratt.me.uk; Rob Jarratt < <mailto:robert.jarratt at ntlworld.com> robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts < <mailto:cctalk at classiccmp.org> cctalk at classiccmp.org> Subject: Re: VAXmate PSU Den ons 8 apr. 2020 kl 00:34 skrev Rob Jarratt via cctalk <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> >: via <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> > supply, a something +V action. the but crowbar - Oh dear! After Brent's question about D19 anode, I realise that the probe was connected to the cathode! I have now done it again with the probe connected to the anode. I have taken two images of the same capture, one at low resolution to show the overall behaviour https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-1.png And one zoomed in to show what happens when the SCR shuts down. https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2. <https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2.png> png The channels are the same as before, namely: Ch1. 555 timer. Ch2. D19 Anode (now corrected as it was previously the cathode!) Ch3. D19 Gate. Ch4. Q1 Source. I got an earlier trace which showed the D19 anode at 9V, which is under the Undervoltage Lockout threshold, but I have not been able to repeat it. I don't fully understand the debate about using the variac. I am not going to debate this either since I know what I have been doing for years and it works perfectly well for me. I have fixed the bigger PSUs in a VAX 11/750 (one broken switch transistor and multiple broken output rectifiers). PSU in NORD-10/S (most carbon composition resistors had gone out of spec). PSUs in many smaller machines as well. I prefer to work in circuits where I can fiddle around without the danger of getting killed all the time. Regardless of use of HV differential probe it can be dangerous. Running it on 50VAC with a protection transformer do expose a lot of problems already and you can poke around safely in the PSU. I have not yet seen a problem that wasn't seen at low voltage, but I expect there could be semiconductors that experience breakdown that occur at lower than specified voltage. However, my measurements appear to suggest that when I use the variac the SCR triggers because of what appears to be a genuine overcurrent detected by R13. I think this is because the duty cycle at low AC input voltages is 50% (rather than about 10% or less as per the trace I have just taken), and I measured 2V across R13, which does seem to be enough to trigger the SCR. When I use 220VAC, the voltage across R13 does rise to 6V, which should also trigger the SCR I think, except that the peak last a lot less and so perhaps the fact that the 6V last for a brief period is insufficient to trigger it? On the issue of duty cycle. If we look at this from the start up perspective rather than the steady state perspective. At startup there are no stored energy in the output filter capacitors. The voltage on the output is thus 0. As soon as the PSU is doing its first switching pulse energy is transfered as the main switch transistor is cutting off. The energy is transfered into the capacitor and into the load. The voltage is starting to increase. The duty cycle generated by the PWM circuitry is in pure relation to the voltage error, i.e. the difference between output voltage and reference voltage. In essence it is a P-regulator. When there are 0 Volt out the duty cycle will be at the maximum. Nothing strange about that. But what is maximum duty cycle? It depends on the circuitry used. The UC3842 can do up to almost 100% duty cycle. However it may be wise to limit duty cycle in a flyback design so that the transformer is not saturated. I am not sure if there is some kind of duty cycle limitation in this circuit though. So if it can handle 50% duty cycle at startup it should be able to handle it at any time. Besides it would be incredible weird to design a circuit to use a 10% duty cycle at its standard operating point and detecting over current at 50%. Then you have much less head room for load and input variations. I am more or less convinced that what you see on the primary side is a result of some kind of fault on the secondary side. A very common problem is short-circuit rectifier diodes on the secondary side (D12, D11, D21, D22, D23, D24). They can be difficult to measure correctly in circuit since the resistance of the secondaries of the transformer is so low. Depending on type you can either desolder them completely or just lift one end of them. My experience is that electrolytic capacitors seldom short circuit. They probably boil and explode instead. Tantalum capacitors often short circuit. Some of them goes into fire other just stay short circuit. So check for tantalum capacitors and try to measure them for short circuit. You have a crowbar on the secondary side. Are you sure that one hasn't triggered? If you still run on variac you can disable the crowbar circuit by removing the SCR and ramp up the voltage slowly to see if that makes any difference. Breaking the feed back loop: R23 seems to be in the feedback path. If you lift it and insert a voltage from a lab supply here you could simulate the output voltage and study the behaviour of the UC3842 for different feed back voltages. You will see that it will stay on max duty cycle up until close to the nominal voltage and over a very small span change to almost no pulse out at all. This is due to the gain of the circuit. /Mattis I have seen the suggestions to study the waveforms at a much higher resolution. What I am doing is setting the overall timebase in the 100ms range so that I can trigger on when the 555 starts to oscillate and capture the whole period of operation until the SCR triggers. I can then zoom in, as can be seen from the trace provided in this email. I hope that is good enough, or am I missing some problem with doing it this way? I would like to follow Mattis's suggestions (and other people have said it too) to break the feedback loop, but it does look difficult to know how best to do it. I also understand Brent's suggestion that the gate spike is just the result of the SCR triggering, rather then the cause of the trigger. I had wondered if that might be the case. Regards Rob

I will repeat it tonight (or tomorrow) with the faster timebase and get the last two cycles. The switching frequency is 100KHz. Thanks Rob From: Mattis Lind <mattislind at gmail.com> Sent: 09 April 2020 08:47 To: rob at jarratt.me.uk; Rob Jarratt <robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org> Subject: VAXmate PSU torsdag 9 april 2020 skrev Rob Jarratt <robert.jarratt at ntlworld.com <mailto:robert.jarratt at ntlworld.com> >: I did use a faster timebase, 10us, triggering on the negative edge of channel 2, and it looked identical to this: https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2… I thought the switching frequency was in the range of 50 kHz. But it might be higher. What would be interesting is to see the last two switch cycles in detail. What is the switching frequency? Admittedly the trace you see was taken on the slower timebase and then zoomed in, but there was no difference, the result was identical. You can loose detail since the sampling rate is a function of sweep frequency/ time base setting. Regards Rob From: Mattis Lind <mattislind at gmail.com <mailto:mattislind at gmail.com> > Sent: 09 April 2020 05:52 To: rob at jarratt.me.uk <mailto:rob at jarratt.me.uk> ; General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> >; Rob Jarratt <robert.jarratt at ntlworld.com <mailto:robert.jarratt at ntlworld.com> > Subject: VAXmate PSU onsdag 8 april 2020 skrev Rob Jarratt <robert.jarratt at ntlworld.com <mailto:robert.jarratt at ntlworld.com> >: I will look at all the suggestions, particularly of a failure on the secondary side. Something must have burned up, because there was a distinct burning smell after the initial failure, although I have never been able to see any physical damage to anything, despite looking many times. Aha. Don?t think I seen you writing about that before, or did you? It might be very hard to find the source some times. Even just a small burn will give quite some smell. Check ALL semiconductors very carefully. But the thing that really puzzles me is that, after correcting the probes to include the D19 anode, there doesn?t seem to be anything that would cause D19 to trigger. Am I reading the trace wrong? It is very hard to tell from the traces what is going on since the resolution is too low. Use a faster timebase. 5 or 10 microseconds. Find out if you can trigger on something that happen only when it stops. Like channel 2 negative slope. /Mattis Thanks Rob From: Mattis Lind < <mailto:mattislind at gmail.com> mattislind at gmail.com> Sent: 08 April 2020 07:42 To: <mailto:rob at jarratt.me.uk> rob at jarratt.me.uk; Rob Jarratt < <mailto:robert.jarratt at ntlworld.com> robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts < <mailto:cctalk at classiccmp.org> cctalk at classiccmp.org> Subject: Re: VAXmate PSU Den ons 8 apr. 2020 kl 00:34 skrev Rob Jarratt via cctalk <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> >: via <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> > supply, a something +V action. the but crowbar - Oh dear! After Brent's question about D19 anode, I realise that the probe was connected to the cathode! I have now done it again with the probe connected to the anode. I have taken two images of the same capture, one at low resolution to show the overall behaviour https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-1.png And one zoomed in to show what happens when the SCR shuts down. https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2. <https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-2.png> png The channels are the same as before, namely: Ch1. 555 timer. Ch2. D19 Anode (now corrected as it was previously the cathode!) Ch3. D19 Gate. Ch4. Q1 Source. I got an earlier trace which showed the D19 anode at 9V, which is under the Undervoltage Lockout threshold, but I have not been able to repeat it. I don't fully understand the debate about using the variac. I am not going to debate this either since I know what I have been doing for years and it works perfectly well for me. I have fixed the bigger PSUs in a VAX 11/750 (one broken switch transistor and multiple broken output rectifiers). PSU in NORD-10/S (most carbon composition resistors had gone out of spec). PSUs in many smaller machines as well. I prefer to work in circuits where I can fiddle around without the danger of getting killed all the time. Regardless of use of HV differential probe it can be dangerous. Running it on 50VAC with a protection transformer do expose a lot of problems already and you can poke around safely in the PSU. I have not yet seen a problem that wasn't seen at low voltage, but I expect there could be semiconductors that experience breakdown that occur at lower than specified voltage. However, my measurements appear to suggest that when I use the variac the SCR triggers because of what appears to be a genuine overcurrent detected by R13. I think this is because the duty cycle at low AC input voltages is 50% (rather than about 10% or less as per the trace I have just taken), and I measured 2V across R13, which does seem to be enough to trigger the SCR. When I use 220VAC, the voltage across R13 does rise to 6V, which should also trigger the SCR I think, except that the peak last a lot less and so perhaps the fact that the 6V last for a brief period is insufficient to trigger it? On the issue of duty cycle. If we look at this from the start up perspective rather than the steady state perspective. At startup there are no stored energy in the output filter capacitors. The voltage on the output is thus 0. As soon as the PSU is doing its first switching pulse energy is transfered as the main switch transistor is cutting off. The energy is transfered into the capacitor and into the load. The voltage is starting to increase. The duty cycle generated by the PWM circuitry is in pure relation to the voltage error, i.e. the difference between output voltage and reference voltage. In essence it is a P-regulator. When there are 0 Volt out the duty cycle will be at the maximum. Nothing strange about that. But what is maximum duty cycle? It depends on the circuitry used. The UC3842 can do up to almost 100% duty cycle. However it may be wise to limit duty cycle in a flyback design so that the transformer is not saturated. I am not sure if there is some kind of duty cycle limitation in this circuit though. So if it can handle 50% duty cycle at startup it should be able to handle it at any time. Besides it would be incredible weird to design a circuit to use a 10% duty cycle at its standard operating point and detecting over current at 50%. Then you have much less head room for load and input variations. I am more or less convinced that what you see on the primary side is a result of some kind of fault on the secondary side. A very common problem is short-circuit rectifier diodes on the secondary side (D12, D11, D21, D22, D23, D24). They can be difficult to measure correctly in circuit since the resistance of the secondaries of the transformer is so low. Depending on type you can either desolder them completely or just lift one end of them. My experience is that electrolytic capacitors seldom short circuit. They probably boil and explode instead. Tantalum capacitors often short circuit. Some of them goes into fire other just stay short circuit. So check for tantalum capacitors and try to measure them for short circuit. You have a crowbar on the secondary side. Are you sure that one hasn't triggered? If you still run on variac you can disable the crowbar circuit by removing the SCR and ramp up the voltage slowly to see if that makes any difference. Breaking the feed back loop: R23 seems to be in the feedback path. If you lift it and insert a voltage from a lab supply here you could simulate the output voltage and study the behaviour of the UC3842 for different feed back voltages. You will see that it will stay on max duty cycle up until close to the nominal voltage and over a very small span change to almost no pulse out at all. This is due to the gain of the circuit. /Mattis I have seen the suggestions to study the waveforms at a much higher resolution. What I am doing is setting the overall timebase in the 100ms range so that I can trigger on when the 555 starts to oscillate and capture the whole period of operation until the SCR triggers. I can then zoom in, as can be seen from the trace provided in this email. I hope that is good enough, or am I missing some problem with doing it this way? I would like to follow Mattis's suggestions (and other people have said it too) to break the feedback loop, but it does look difficult to know how best to do it. I also understand Brent's suggestion that the gate spike is just the result of the SCR triggering, rather then the cause of the trigger. I had wondered if that might be the case. Regards Rob

Here is a trace taken using a 5us timebase. https://rjarratt.files.wordpress.com/2020/04/primary-side-shutdown-detail-h… As a reminder the probes are connected as follows: Ch1. 555 timer. Ch2. D19 Anode Ch3. D19 Gate. Ch4. Q1 Source. It doesn't look like there are any glitches on the D19 gate. I don't think it is clear why D19 is triggering, because all the previous oscillations are about the same. I did check Q2 before, but perhaps I should check again. I have checked for shorts on the actual outputs, but there don't seem to be any. I have checked one of those rectifiers, I think one of my next tasks is to desolder all of them and check them.

I have found the problem and a possible other problem too. I have discovered that D24 is shorted (no visible damage), although its twin D23 is not. I will replace both. I know it is shorted because I lifted one end. I have also checked the rectifiers, D11 and D12. I am not sure if there is a possible problem here too. D12 is an MBR3045PT. It tests correctly as a common cathode diode network. However, the forward voltage seems to be 0.19V. The datasheet (https://pdf1.alldatasheet.com/datasheet-pdf/view/53622/FAIRCHILD/MBR3045PT.…) would suggest it should be 0.76V at room temperature. I can see no physical damage to it though. D11 is marked PHS 2402. It too tests correctly as a common cathode diode network, with a forward voltage of 0.43V. I can't find a datasheet for it though, but I believe the equivalent is MUR1610CT (https://pdf1.alldatasheet.com/datasheet-pdf/view/82128/ONSEMI/MUR1610CT.html) and I think the forward voltage for that should be 0.9V. Again no physical damage that I can see. Is it possible both rectifiers are marginal? Or at least D12? Should I replace one or both? Thanks Rob

I took a break and while I was doing that I thought the same thing, I am going to see if I can see a reason for the failure. I did wonder if the tester?s current would be too low to show the nominal forward voltage on D11 and D12. I won?t change those. Thanks Rob From: Mattis Lind <mattislind at gmail.com> Sent: 10 April 2020 15:16 To: rob at jarratt.me.uk Cc: General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org>; Matt Burke <matt at 9track.net> Subject: Re: VAXmate PSU I have found the problem and a possible other problem too. But that is great! I have discovered that D24 is shorted (no visible damage), although its twin D23 is not. I will replace both. I know it is shorted because I lifted one end. Well, sometime there are no visible damage but still a smell. I have also checked the rectifiers, D11 and D12. I am not sure if there is a possible problem here too. D12 is an MBR3045PT. It tests correctly as a common cathode diode network. However, the forward voltage seems to be 0.19V. The datasheet (https://pdf1.alldatasheet.com/datasheet-pdf/view/53622/FAIRCHILD/MBR3045PT.…) would suggest it should be 0.76V at room temperature. I can see no physical damage to it though. The point is that the measurement in the data sheet is with 20A forward current. You meter perhaps give 10 mA forward current. This is a Schottky rectifier it has very low Vf. 0.19 is completely normal. D11 is marked PHS 2402. It too tests correctly as a common cathode diode network, with a forward voltage of 0.43V. I can't find a datasheet for it though, but I believe the equivalent is MUR1610CT (https://pdf1.alldatasheet.com/datasheet-pdf/view/82128/ONSEMI/MUR1610CT.html) and I think the forward voltage for that should be 0.9V. Again no physical damage that I can see. Is it possible both rectifiers are marginal? Or at least D12? Should I replace one or both? No. At this point replace just D24 and possibly D23. Actually I don't think it is necessary. But to be on the safe side it might be a good idea. But the question is why it failed. Sometimes they fail because of just nothing. But it could be something in the network following it that had made it fail. Check that there are no short circuit in the LM317 regulator. Check output capacitors in this area. ESR and value. /Mattis Thanks Rob

Thanks Eric, I had a private reply that said pretty much the same thing. Last night I replaced the shorted diode and another electrolytic capacitor and the PSU started up on the bench, powering a load board and an RD53 hard disk. Ripple was good except on the -9V output, but that output doesn?t seem to have any capacitors after the final transformer, so I think it is OK. I need to replace a few marginal capacitors on the video module now, while I have the machine in bits. Then I will put it all back together to see that it still works. I have a lingering worry that the failed diode was on the +28V supply to the video module and that there might be a fault on the video module, but it doesn?t appear to present a short circuit, so I am hopeful. I will do a blog post once it is all back up and running. Thanks Rob From: Eric Smith <spacewar at gmail.com> Sent: 16 April 2020 00:05 To: rob at jarratt.me.uk; Rob Jarratt <robert.jarratt at ntlworld.com>; General Discussion: On-Topic and Off-Topic Posts <cctalk at classiccmp.org> Subject: Re: VAXmate PSU On Fri, Apr 10, 2020 at 10:14 AM Rob Jarratt via cctalk <cctalk at classiccmp.org <mailto:cctalk at classiccmp.org> > wrote: D12 is an MBR3045PT. It tests correctly as a common cathode diode network. However, the forward voltage seems to be 0.19V. The datasheet (https://pdf1.alldatasheet.com/datasheet-pdf/view/53622/FAIRCHILD/MBR3045PT.…) would suggest it should be 0.76V at room temperature. I can see no physical damage to it though. Even when they are good, the forward drop of a diode can be FAR lower than the typical forward drop (around 0.7 for normal silicon diodes) if you're putting significantly less that the rated current through it. 0.19V forward drop is only slightly lower than typical characteristic at e.g. 10mA current. See figure 3. I've seen more than enough variation of diodes from typical curves for that alone to convince me that the diode is bad (though obviously it may be). The real test is how much it conducts in reverse. All diodes will pass a small amount of reverse current. This one shouldn't pass more than 1 mA in the reverse direction at room temperature, even with near the rated reverse voltage (45V) applied.

Den s?n 5 apr. 2020 kl 23:53 skrev Rob Jarratt via cctalk < cctalk at classiccmp.org>: When you use a variac the developed voltage on the secondary side will be much lower due to the turn ratio of the transformer. You measured 0.4V on the 5V line if I understand you correctly. That means approx 1/12 of nominal output. That would then lead to 1/12 current on the secondary side and 1/144th of the power developed. With this little power developed on the secondary side there is no chance you can exceed the power transistor ratings. I do this all the time. I measure transistor current when doing this and there are not a ten fold or so increase in current as Brent claim. By using a variac and you can measure and study waveforms and detect problems in driving circuits without killing the main switch transistor. You can study actions of crowbar functions. There are two different basic variants for SMPS mode PSUs. Forward or Flyback. Forward are normally used in higher power designs while flyback is lower power. A forward mode PSU is like creating a square wave AC current and feeding it to a transformer. In forward mode the duty cycle cannot be more than 50% since the core need to reset before next cycle. The output will be a in relation to the input by the turn ratio of the transformer. A forward converter need a way to reset the transformer, either by various diodes on the primary diode or using various bridge topologies for the switching system. There are also filtering inductors on the secondary side. However your PSU look very much like a flyback design. It means that the transformer has an airgap. It is actually both and inductor and transformer. A flyback converter stores energy in the inductor when the transistor is turned on and then when turned off the energy is transfered to the load. When the transistor is switched on the current will increase linearly in the primary winding. This will continue until the transistor is switched off or the core is saturated. When saturated the current will rise sharply. This also means that the duty cycle can be almost 100% only limited by the design of the transformer so that it does not saturate. Saturation is a condition you don't want to have. What could be interesting is to study the waveform of the drain current in more detail. You need to set timebase of your oscilloscope to a much shorter time span to in detail study the increase of current in the drain. The UC3842 does allow up to 100% duty cycle while the other cousins in the same family, UC3844 and UC3845, only go up to 50%, thus made for forward converters. It can be interesting to investigate if the feed back circuits works as expected. Usually one can inject lab supply voltages on the outputs and find out if the control circuitry reacts properly. The problem with your supply is that it is fairly complicated on the secondary side. While reading the schematic it is not immediately clear that the op-amps will be operating properly by feeding in +5V and +12 V backwards, unfortunately. Then another option is to cut open the feed back path and inject a feedback voltage from a labsupply at the input of the UC3842. You can then vary it while there are supply voltage to the UC3842 and you can then monitor the duty cycle of the UC3842. If you cut away the secondary side output stage you have to bear in mind that this is most likely a flyback design. Without any load at the output of the transformer the energy stored in the magnetic field has no where to go. You could get very high voltages on the drain. I suggest that you have a load resistance over each output of the transformer. By doing this together with breaking up the feed back loop as described above you could monitor how the primary side is doing its switching. You should not try to do this without a variac. While having the output stages cut away from the transformer you could also test the output stages by powering them from the transformer side by a bench supply. What happens if you turn up the voltage slowly? Will the crowbar trigger or not. Is a proper feed back voltage generated. A lot of things to test, but one common note is that you need to study your waveforms on a cycle basis not 100 ms/division. More like 10 us/division. Good luck! /Mattis

I like this idea, I will have to see how best to do that. Regards Rob