Hi Suzyj,
I have been following your very interesting cascading diamond buffers thread.
It was there that I saw your comment on some spikes on the LTP stage being good for linearity, or I understood it that way, and raised my curiosity about such a thing. Did I get it right?
Are you certain about the SST404 being dual? Calogic lists it's single:
Calogic
Mouser says it's obsolete. Would the 2N4341 work? It's a single, but Mouser stocks it.
Price is high, so I think an LSK389 from Linear Systems might be a better idea.
The U404 is dual indeed, but Mouser also says it's obsolete. Future Electronics has it? Any idea of price?
I have been following your very interesting cascading diamond buffers thread.
It was there that I saw your comment on some spikes on the LTP stage being good for linearity, or I understood it that way, and raised my curiosity about such a thing. Did I get it right?
Are you certain about the SST404 being dual? Calogic lists it's single:
Calogic
Mouser says it's obsolete. Would the 2N4341 work? It's a single, but Mouser stocks it.
Price is high, so I think an LSK389 from Linear Systems might be a better idea.
The U404 is dual indeed, but Mouser also says it's obsolete. Future Electronics has it? Any idea of price?
On the basis of availability in Australia it would seem perhaps try locally (Brazil) at Contact Us
- Future Electronics
Last edited:
First of all, we are talking of the 5M21, right? Because those notches of the L10 I could not tame. On the 5M21, using the circuit I uploaded, the notches are gone.
Same thing for the small cap going from VAS out to LTP negative input. That could soften a spike in the square wave. Do the same: cut C212 on the square simulation I uploaded and watch the result.
OTOS, if you do that on the THD simulation, you will see that THD goes even lower. So I compromised on that capacitor value to improve square wave response for a little worst THD. Will that be sonically audible?
In any case, as I stated, those "errors" might be the designer's intention for a specific sound signature. So I think they should be tried afterwards on the copper side. If they audibly improve audio quality, then they might be part of the final pcb. Don't you think?
Just to check again, I went back to the original schematic, unchanged except for the active parts which are unavailable. Playing with those caps you mentioned, C207 and C301, do not eliminate the glitches in the LTP out sinewave, whatever the frequency. In fact things get better when you bypass the drivers emitter resistors with a 10uF cap.
Good point. I wonder how you can simulate slew rate in an amp with LTSpice, to see if the SR changes with different output bipolars. I do know things change when you try different types, particularly THD. Perhaps a separate pcb board should be built for different output transistors. Particularly metal TO-3 types, which the original amp used.
I wish the LTP square wave was better. It gets fine after the VAS and at the output.
Now what is that about C207? If it was 33pF somewhere it was a mistake. But I just simulated the original circuit using 33pF on 207, and it didn't change anything on the LTP.
SST404: SST404-LF | SST404 Series 50 V 10 mA N-Channel JFET Monolithic Dual - SOIC-8 | CALOGIC
- Future Electronics
U404 (TO71): U404 TO-71 6L | Linear Integrated Systems | Trendsetter Electronics
U401 (better matched U404): U401 | U401 Series 50 V 10 mA Dual N-Channel JFET Switch - TO-71 | CALOGIC
- Future Electronics
Also worthwhile looking at the LSK489: Trendsetter Electronics - Electronic Components Distributor
- Future Electronics
U404 (TO71): U404 TO-71 6L | Linear Integrated Systems | Trendsetter Electronics
U401 (better matched U404): U401 | U401 Series 50 V 10 mA Dual N-Channel JFET Switch - TO-71 | CALOGIC
- Future Electronics
Also worthwhile looking at the LSK489: Trendsetter Electronics - Electronic Components Distributor
Last edited:
The U404 & SST404 are the same transistors, just packaged differently.
Here's the model I use for the U/SST404:
.model U404 NJF(Beta=1.577m Betatce=-.5 Rd=1 Rs=1 Lambda=10m Vto=-1.316 Vtotc=-2.5m Is=19.73f Isr=191.3f N=1 Nr=2 Xti=3 Alpha=68.56u Vk=212.2 Cgd=5.6p M=.3916 Pb=.5 Fc=.5 Cgs=6.044p Kf=4.592E-18 Af=1 mfg=Vishay)
The noise spec given for the U/SST404 is at 10Hz, where flicker noise is predominant. In a real-life application with the SST404 I get around 5nV/sqrtHz input referred at 1KHz, so they're reasonably good. I've modelled the same amp with the LSK489 and didn't get any improvement in noise, though that's likely a combination of poor models and non-optimised input. I haven't tried the real thing as the LSK part has a different pinout so I'd need a board spin. For the application (power amp input stage) I'm happy enough with 5nV/sqrtHz.
Here's the model I use for the U/SST404:
.model U404 NJF(Beta=1.577m Betatce=-.5 Rd=1 Rs=1 Lambda=10m Vto=-1.316 Vtotc=-2.5m Is=19.73f Isr=191.3f N=1 Nr=2 Xti=3 Alpha=68.56u Vk=212.2 Cgd=5.6p M=.3916 Pb=.5 Fc=.5 Cgs=6.044p Kf=4.592E-18 Af=1 mfg=Vishay)
The noise spec given for the U/SST404 is at 10Hz, where flicker noise is predominant. In a real-life application with the SST404 I get around 5nV/sqrtHz input referred at 1KHz, so they're reasonably good. I've modelled the same amp with the LSK489 and didn't get any improvement in noise, though that's likely a combination of poor models and non-optimised input. I haven't tried the real thing as the LSK part has a different pinout so I'd need a board spin. For the application (power amp input stage) I'm happy enough with 5nV/sqrtHz.
This wish prompted me to look a bit deeper into your circuit I had assumed from the fact it was stable at high power with 2uF//8R it would be satisfactory at all signal input levels with 10kHz square wave with a variety of capacitor values - smaller ones excite higher hf resonances.
As your actual hardware stands you have C212 in place but not the 270 pF capacitors I found a patch of oscillation at the bottom of the falling wave at lower outputs - so you are not quite in the clear with this.
If you leave C212 and fit the 270 pF capacitors what should be the flat tops and bottoms of the square wave are warped.
The situation is not improved if C212 is removed and the 270 pF capacitors are fitted either.
C212 leads feedback from the Vas back to the inverting input bypassing the inevitable delays in the output stage while C203 leads the delayed signal from the output back to the same inverting input which sums the two together. The best analogy to describe this is like the echo from a public address system in a hall - hard to understand.
My answer to the problem is to delete C203 and double C212 in value to 4.7 pF.
My modified simulation and a screen shot of the square wave with various parallel caps are attached.
Attachments
Great help, mjona! Particularly because you are using ways I didn't know.
I mean the .step parameters, which generate several specs and curves at the same time.
Is there a way to separate each curve or see each one separately? On the THD sim I have the THD results, but on the sqw one I can't see the different curves, like the image you sent.
Another good thing would be to get a family of curves to see THD and sqw results using different input levels or frequencies. Is that possible?
It's a pity there are no models for the original transistors, which would allow spec comparisons or some idea of it.
I mean the .step parameters, which generate several specs and curves at the same time.
Is there a way to separate each curve or see each one separately? On the THD sim I have the THD results, but on the sqw one I can't see the different curves, like the image you sent.
Another good thing would be to get a family of curves to see THD and sqw results using different input levels or frequencies. Is that possible?
It's a pity there are no models for the original transistors, which would allow spec comparisons or some idea of it.
If only one point of the circuit is displayed the multiple plots show in one shot distinguished by different colours - first screen shot.
The second shot shows that the waveform at the feedback take-off point is not disrupted by the load with the various parallel capacitors - more on this later.
If you look at the rising slope in this shot it is possible to get an indication of slew rate at various points.
Settling time is that from zero volts to flat at the top of the square wave - there are no overshoots or ripple at top or bottom in this simulation.
There is a little rounding on the corner near the top. Laying the cursor across the flat part of the top the reading shows on the y scale at the left in the bar at the bottom of the screen as 28.6 volts.
Moving the cursor to the point near the rising corner at the top avoiding the rounded portion shows the x time scale as 1.83 us. Dividing this into 28.6 volts gives 15.6 V/us - a a bit rough but close enough.
In a domestic setting an output power of 1 watt can be loud enough so I looked at the x and y axes at 8 volts and read the time off as 0.43 us equal to 18 V/us.
Rise time is inversely proportionate to rise time - as a useful measure 10us equates to a frequency of 100 kHz.
By rough calculation 1.83 us corresponds to 546 kHz and 0.43 us to 2.3 MHz.
The 10 kHz square wave test is to check loop stability.
The natural increase of a coil impedance with increasing frequency in combination with a load including any capacitor will generate more resonance. The displays in the first screen shot represent frequencies outside the audio range - they are not to be taken as signs of any problem therein.
Besides deleting C203 and increasing C212 to 4.7pF, is there anything I could or should do to improve SqW response on this amp?
I am not sure in what order should I go modding things on the actual prototype, starting from the original circuit. Some solve the glitch in the LTP sinusoidal, others improve things on the SqW response.
I am not sure in what order should I go modding things on the actual prototype, starting from the original circuit. Some solve the glitch in the LTP sinusoidal, others improve things on the SqW response.
The first step would be to do a square wave test on your prototype as built with a few of the lowest value capacitors used in the simulation// 8R test load with a level input signal - it was at low level that I saw the signs of instability in my simulations.
Next photo the screen shot and compare the image with that of your .raw file result for the same test conditions.
When making changes do this one step at a time and with each step compare the channel that has been modified with it's unmodified partner.
You can step left or right with either channel as you go forward and backward in direction if you meet a with a problem.
With regard to increasing C212 to 4.7pf that was the lowest effective value than served as a cure.
There could be a need to experiment to find the needed value due to the properties of circuit board material wiring and layout etc.
You might find one channel needs a different capacitor value from it's partner.
I have looked at ways to improve the symmetry between rising and falling slope settling times - which involves changing some other capacitor values - such could be affected depending on whether or not 4.7 pF is the right value for C212.
Mjona,
How do you set the level for square wave tests?
Pity that most tests you suggest I will only be able to listen to, because I don't have the necessary instruments.
In the meantime I've looked in the L10 simulation, as I got to lower THD on another Luxman amp simulation by increasing bias current.
So I did the same on the L10 to see if I could cure the sinusoidal glitches on the LTP. To get there I had to triple the current and add a bypass cap over the bias transistor CE. That seemed to cure the glitch and lowered THD at 20K. BTW, the glitches only happened at higher frequencies.
If you feel like it, please carry on your SqW tests on the L10 too.
The L10 is very similar to the 5M20, minus the pre-driver/buffer. But the remedies I applied on the 5M20 for the glitches had not worked on the L10.
The price to pay on the L10 is the watts the output bipolars have to dissipate: 5W each. But I think that's within their specs and cured with good heatsinks, don't you think?
How do you set the level for square wave tests?
Pity that most tests you suggest I will only be able to listen to, because I don't have the necessary instruments.
In the meantime I've looked in the L10 simulation, as I got to lower THD on another Luxman amp simulation by increasing bias current.
So I did the same on the L10 to see if I could cure the sinusoidal glitches on the LTP. To get there I had to triple the current and add a bypass cap over the bias transistor CE. That seemed to cure the glitch and lowered THD at 20K. BTW, the glitches only happened at higher frequencies.
If you feel like it, please carry on your SqW tests on the L10 too.
The L10 is very similar to the 5M20, minus the pre-driver/buffer. But the remedies I applied on the 5M20 for the glitches had not worked on the L10.
The price to pay on the L10 is the watts the output bipolars have to dissipate: 5W each. But I think that's within their specs and cured with good heatsinks, don't you think?
Sorry, Ronaldo. I completely missed your suggestions. I probably saw the Brazilian flag and I thought it was my comment.
Of course: buying the amp on eBay, as you well know, is out of the question from Brazil. Also you don't know what repairs were made until you get it, or what it's wrong with it.
Designing the pcb based on the original one is probably what I will do when I get to that stage. Finding the original bipolars is impossible, as you also may know. Any similar names parts you find on eBay will be 98% fakes.
Of course that option 3 is the best to try. I will remain as close as possible to the original amp, and depend on listening tests, compared to the original amp and on its own, how it sounds. For that I will depend on my listening experience as professional audio person.
Of course I would love to see your HTForum 10-year old project, if you give me the URL. Or even the Mark Levinson amp you liked better.
The main reason why I got interested in this amp is because I got to know its audio signature, which many will say it's something that doesn't exist. But my experience in audio showed me a difference thing there.
Again, please forgive my not having answered your comments at the right time. You're a good friend and I wouldn't do that.
I think it's time to update on further developments this project could get to, thanks to Michael Jonassen (Mjona) from New Zealand and Giancarlo Foli from Italy.
Intensive LTSpice modelling was applied to the original version particularly using Tian Probe sims to look at stabilization margins.
If there's anyone around interested, I can upload the different asc files for you to see, sim and comment.
The project seems now ready for prototype building, with what seem to be better specs and stability.
The 5M21 proved better than the L10 in the simulations, much more stable too. The 5M21 also simulates with much better results than all the family of "brothers", like the L10, M12 or B12.
Intensive LTSpice modelling was applied to the original version particularly using Tian Probe sims to look at stabilization margins.
If there's anyone around interested, I can upload the different asc files for you to see, sim and comment.
The project seems now ready for prototype building, with what seem to be better specs and stability.
The 5M21 proved better than the L10 in the simulations, much more stable too. The 5M21 also simulates with much better results than all the family of "brothers", like the L10, M12 or B12.
Here are the latest asc files for Shadow 5M21, the clone that is not a clone.
My major concern was to keep the original architecture as much as possible, so no major changes there.
The simulation specs seem even better than the original design, but real life parts may probably shatter that finding in the actual prototype.
The three major areas that were worked were THD, square wave and Tian Probe. Please feel free to play with the simulations to get other results.
The original gain ratio of x32 was kept, but the resistors pair was changed, from 47K/1K5 to 32K/1K. Using the latter improves the gain margin from 12dB to 16dB, and the phase margins from 58 degrees to 62dB.
THD gets a little bit worst with this new pair, but the difference is too small to care about.
My major concern was to keep the original architecture as much as possible, so no major changes there.
The simulation specs seem even better than the original design, but real life parts may probably shatter that finding in the actual prototype.
The three major areas that were worked were THD, square wave and Tian Probe. Please feel free to play with the simulations to get other results.
The original gain ratio of x32 was kept, but the resistors pair was changed, from 47K/1K5 to 32K/1K. Using the latter improves the gain margin from 12dB to 16dB, and the phase margins from 58 degrees to 62dB.
THD gets a little bit worst with this new pair, but the difference is too small to care about.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.