Congratulations @r580808 , on building and bench testing a pair of "Yes It Can Drive an F4" boards! Very nice job.
Even before you complete the experiment suggested by @thimios in post #139, I'd like to mention a few facts about your square-wave-in, big-slope-and-not-very-square-wave out at 50 Hertz, which is shown in your final photo of post #136.
The input square wave in your photo has a flat top and a flat bottom, while the output slopes diagonally. The slope is expected and entirely normal. It happens because the AC coupling capacitor "C1" (2.2 microfarad polypropylene film) at the input, is sloooooowly discharged through bias resistor R4. Usually the sloooooow discharge is completely negligible; it only matters with signals that resemble DC --- like the flat tops and flat bottoms of looooooow frequency square waves.
Not only is it visible in real life amplifiers; this phenomenon also appears in circuit simulations. I've circled the part of YICDAF4 which creates this slopey-ness, with a red circle on the schematic. R1=330K doesn't actually do anything because it's in parallel with a voltage source, but I left it in the simulation anyway. Just in case skeptical people feel it MUST be included in the simulation.
Figure 2 shows the question I asked LTSPICE. "When I attach a square wave to these two resistors and one capacitor, what do the output look like?"
Figure 3 is LTSPICE's answer. The parts of the input square wave which resemble DC, namely the flat top and flat bottom, allow R4 to slooooooowly discharge C1 and the output slopes.
Figure 4 is the answer to a different question: "What if I increase C1, from 2.2uF to 22uF? Does the slope change?" LTSPICE's answer is Yes. The slope is reduced by a factor of ten. When the capacitance increases by 10x , the slope decreases by 10x. If that's very important to you, please feel free to find a 22uF polypropylene film capacitor which fits the PCB footprint, and replace C1. Be sure to get one whose DC voltage rating is comfortably greater than 30V.
I realize this is not the main thrust of @r580808 's inquiry, but rather a side quest. When the ground clip placement experiment has been completed, we'll deal with the other scope photos.
Even before you complete the experiment suggested by @thimios in post #139, I'd like to mention a few facts about your square-wave-in, big-slope-and-not-very-square-wave out at 50 Hertz, which is shown in your final photo of post #136.
The input square wave in your photo has a flat top and a flat bottom, while the output slopes diagonally. The slope is expected and entirely normal. It happens because the AC coupling capacitor "C1" (2.2 microfarad polypropylene film) at the input, is sloooooowly discharged through bias resistor R4. Usually the sloooooow discharge is completely negligible; it only matters with signals that resemble DC --- like the flat tops and flat bottoms of looooooow frequency square waves.
Not only is it visible in real life amplifiers; this phenomenon also appears in circuit simulations. I've circled the part of YICDAF4 which creates this slopey-ness, with a red circle on the schematic. R1=330K doesn't actually do anything because it's in parallel with a voltage source, but I left it in the simulation anyway. Just in case skeptical people feel it MUST be included in the simulation.
Figure 2 shows the question I asked LTSPICE. "When I attach a square wave to these two resistors and one capacitor, what do the output look like?"
Figure 3 is LTSPICE's answer. The parts of the input square wave which resemble DC, namely the flat top and flat bottom, allow R4 to slooooooowly discharge C1 and the output slopes.
Figure 4 is the answer to a different question: "What if I increase C1, from 2.2uF to 22uF? Does the slope change?" LTSPICE's answer is Yes. The slope is reduced by a factor of ten. When the capacitance increases by 10x , the slope decreases by 10x. If that's very important to you, please feel free to find a 22uF polypropylene film capacitor which fits the PCB footprint, and replace C1. Be sure to get one whose DC voltage rating is comfortably greater than 30V.
I realize this is not the main thrust of @r580808 's inquiry, but rather a side quest. When the ground clip placement experiment has been completed, we'll deal with the other scope photos.
Attachments
This is the waveform I got after inputting a square wave signal (10kHz) when I made the preamplifier (B1 with Korg Nutube, B1K). Please refer to it.
I've attached a plot of Open Loop Gain versus frequency, for Yes It Can Drive An F4 (abbrev YICDAF4) with inductor L1 present (green trace) and again with inductor L1 removed (red trace). It illustrates some of the ideas discussed back in post #1 of this thread:
Pause for a moment and think about that. Adding one small and inexpensive component, increased the gain bandwidth product by a factor of twenty. Doesn't that sound almost too good to be true? Surely there's got to be some kind of a tradeoff, or hidden cost, or unhappy side effect? As Milton Friedman and Robert Heinlein preach: There Aint No Such Thing As A Free Lunch . . . . however this inductor may appear to be a counterexample (?)
Indeed TANSTAAFL , you are right to be skeptical. The hidden tradeoff, the downside of adding the inductor, is an increase in the amplifier's Settling Time. Scott Wurcer pointed this out, here on these Forums, more than 17 years ago:
In (this post), Scott Wurcer wrote
The analysis of this phenomena requires a fairly deep dive into control systems theory, but fortunately the kind people at Analog Devices have summarized it very nicely in their application note AN359, which I attach below. In particular, Figure 5 of the app note (snipped and attached), shows that when the open loop gain hinges upward near the gain crossover frequency (labeled "m < 1"), the settling time increases and the output waveform has transient overshoot.
And that's exactly what we have in YICDAF4: an open loop gain curve that hinges upward (see green trace on the OLG vs freq curve), and a transient overshoot. AN359 predicted it beautifully.
It's worth considering: is this kind of overshoot / settling time , actually detrimental? Is it anything to worry about? After all, the JE990 discrete opamp has been manufactured and sold in the professional audio market for more than 40 years. JE990 contains Long Tailed Pair inductors which produce "nasty settling time", yet the pro audio folks keep buying JE990s and using them. Maybe settling time is crucially important for D-to-A uses, but perhaps not so important for continuous time, continuous signal, linear circuits. Maybe the pro audio people who buy and use JE990s (and Burwen's ADI121 before then), aren't complete dunces?
Thanks to Analog Devices we now have a theoretical understanding of the settling time and overshoot of YICDAF4. It's a side effect of the 20X increase in gain bandwidth product (and 18 dB more feedback where distortion is worst: at 20 kHz). I also did some lab measurements of my own, on YICDAF4 boards with the called-for inductor, and again with modified inductors. I'll post those results after I've gathered everything together.
Pause for a moment and think about that. Adding one small and inexpensive component, increased the gain bandwidth product by a factor of twenty. Doesn't that sound almost too good to be true? Surely there's got to be some kind of a tradeoff, or hidden cost, or unhappy side effect? As Milton Friedman and Robert Heinlein preach: There Aint No Such Thing As A Free Lunch . . . . however this inductor may appear to be a counterexample (?)
Indeed TANSTAAFL , you are right to be skeptical. The hidden tradeoff, the downside of adding the inductor, is an increase in the amplifier's Settling Time. Scott Wurcer pointed this out, here on these Forums, more than 17 years ago:
In (this post), Scott Wurcer wrote
The analysis of this phenomena requires a fairly deep dive into control systems theory, but fortunately the kind people at Analog Devices have summarized it very nicely in their application note AN359, which I attach below. In particular, Figure 5 of the app note (snipped and attached), shows that when the open loop gain hinges upward near the gain crossover frequency (labeled "m < 1"), the settling time increases and the output waveform has transient overshoot.
And that's exactly what we have in YICDAF4: an open loop gain curve that hinges upward (see green trace on the OLG vs freq curve), and a transient overshoot. AN359 predicted it beautifully.
It's worth considering: is this kind of overshoot / settling time , actually detrimental? Is it anything to worry about? After all, the JE990 discrete opamp has been manufactured and sold in the professional audio market for more than 40 years. JE990 contains Long Tailed Pair inductors which produce "nasty settling time", yet the pro audio folks keep buying JE990s and using them. Maybe settling time is crucially important for D-to-A uses, but perhaps not so important for continuous time, continuous signal, linear circuits. Maybe the pro audio people who buy and use JE990s (and Burwen's ADI121 before then), aren't complete dunces?
Thanks to Analog Devices we now have a theoretical understanding of the settling time and overshoot of YICDAF4. It's a side effect of the 20X increase in gain bandwidth product (and 18 dB more feedback where distortion is worst: at 20 kHz). I also did some lab measurements of my own, on YICDAF4 boards with the called-for inductor, and again with modified inductors. I'll post those results after I've gathered everything together.
Attachments
Thanks... for the post.... reading through the App note.
The gain is affected below 100Khz... does this mean that the over shoot only happens under 100Khz?
From a digital point of view, our frequencies are much higher so perhaps we are not affected by the overshoot at lower frequencies? Also in digital systems we normally trigger at a level transition, so, so long as their is no post-ringing we are not so concerned about the over shoot. Undershoot might affect -delay- the rise time... But it looks like the inductor doesn't affect the circuit for a digital application.
I might be wrong.... as usual.. but that's why I see... but it seems that digital and analog signal processing are so different.... and for that matter, video processing is also much higher. So I don't know that it would be affected.
Also, again still looking through the stuff.. but... in analog circuits the frequencies are much lower... what are the frequency components of the overshoot? Meaning a short pulse might have very high frequency components and a very short time that filtered in audio systems? Speakers for example and perhaps transformers?
Nice post, btw.
Last edited:
Thanks for the explanation. Is it possible to add a diode to eliminate the side effects of the inductor? As this YouTube video explains.
If this were a DC motor drive attached to a switch, sure. The circuit works in the video because the springy thing's energy (coil) needs a place to go when the DC current stops, and the diode acts as a one-way relief valve.
However this is an (AC) amplifier, which is using the "springiness" of the coil to boost the gain and voltage swing in a way that has a lot of advantages, with the side effect of a little bit of overshoot. I.E., it's part of the circuit as designed.
However this is an (AC) amplifier, which is using the "springiness" of the coil to boost the gain and voltage swing in a way that has a lot of advantages, with the side effect of a little bit of overshoot. I.E., it's part of the circuit as designed.
That’s super easy to test… remove one leg of the inductor and try it. The good news is it’s OK to do, as mentioned in post one -
Mouser Greece (website gr.mouser.com) stocks two of the three C4 candidates mentioned in post # 125 of this thread. The first candidate has 380 pieces in stock today at gr.mouser and the third candidate has 5,594 pieces in stock today at gr.mouser. Both of them are X7R and both of them are 2.2uF -- exactly as the circuit design & Detailed Parts List calls for.
DigiKey Greece (website www.digikey.gr) has all three of the post #125 candidates in stock today. The first one has 24 pieces in stock at digikey.gr, the second one has 2,493 pieces in stock at digikey.gr, and the third one has 2,345 pieces in stock at digikey.gr.
gr.mouser.com has 25 different models of 2.2K resistor in stock today (here is the list) and digikey.gr has 34 different models of 2.2K resistor in stock today (a link to their table) so I don't think you need to substitute a different resistor value for R18.
DigiKey Greece (website www.digikey.gr) has all three of the post #125 candidates in stock today. The first one has 24 pieces in stock at digikey.gr, the second one has 2,493 pieces in stock at digikey.gr, and the third one has 2,345 pieces in stock at digikey.gr.
gr.mouser.com has 25 different models of 2.2K resistor in stock today (here is the list) and digikey.gr has 34 different models of 2.2K resistor in stock today (a link to their table) so I don't think you need to substitute a different resistor value for R18.
Provides the experimental results after removing the inductor, the input signal is a square wave of 1 volt. To me, it feels better than adding an inductor, and the output is still 20 volts.
The cost is 18 dB less open loop gain at 20 kHz , i.e. 18 dB (8X) less feedback for distortion reduction.
Richard Burwen, one of the founders of Analog Devices, preferred to include the inductor. You, on the other hand, are free to leave it out if you wish.
Richard Burwen, one of the founders of Analog Devices, preferred to include the inductor. You, on the other hand, are free to leave it out if you wish.
I have not tried the YICDAF4 without the inductors, however, I can say building it as designed, the F4 with these front ends continues to impress me every single time I play some tunes. Just sounds awesome.
Thanks again Mark for sharing this design with the community. Your generosity is an amazing gift and inspiration!
Thanks again Mark for sharing this design with the community. Your generosity is an amazing gift and inspiration!