However, it is possible to slightly increase the capacitance of capacitor C2. It would reduce the distortion, which is mostly above 50kHz. Still, a real object and IMD distortion measurements would be needed. However, even the potential risk of clipping for ultrasound frequencies is unlikely because (spectrogram of music):
I measured the impedance of a three-way speaker (Z=8Ω, including 3m of connecting cable) with an impedance meter, and here are the measurement results for the series-connected components model Rs,Cs,LsR_s, C_s, L_sRs,Cs,Ls, and the parallel-connected components model Rp,Cp,LpR_p, C_p, L_pRp,Cp,Lp:
100Hz: Z= 12.9Ω, Rs=10.7Ω, Cs=217nF, Ls=-11.6mH, Rp=15.7Ω, Cp=69.6μF, Lp=-36.4mH
1kHz: Z= 21.8Ω, Rs=21.3Ω, Cs=-35.2μF, Ls=719uH, Rp=22.3Ω, Cp=-1.51μF, Lp=16.7mH
10kHz: Z= 14.2Ω, Rs=14.2Ω, Cs=87.5μF, Ls=-2,90μH, Rp=14.2Ω, Cp=14.4nF, Lp=-17.6mH
50kHz: Z= 18.5Ω, Rs=15.7Ω, Cs=-326nF, Ls=31.1μH, Rp=21.7Ω, Cp=-91.2nF, Lp=111μH
100kHz: Z= 25.5Ω, Rs=18.5Ω, Cs=-90.4nF, Ls=28μH, Rp=35.2Ω, Cp=43.1nF, Lp=58.8μH
In the frequency range of interest, i.e., where we observe ripples in the frequency response, the inductive component of the impedance is less than 100 μH. For the series model, it is approximately 30 μH, and for the parallel model, we have an inductance of about 60 μH in parallel with a capacitance of approximately 40 nF. My meter is not capable of measuring impedance at higher frequencies, but there is a visible trend of a slight decrease in the series inductance component and a much more significant decrease for the parallel inductance.
This impedance meter is a "substitute" RLC bridge utilizing the Cyrustek ES168/ES169 chipset, whose description, along with the measurement method and calculation algorithms, is detailed here: http://www.cyrustek.com.tw/wp-content/uploads/ES168.pdf
100Hz: Z= 12.9Ω, Rs=10.7Ω, Cs=217nF, Ls=-11.6mH, Rp=15.7Ω, Cp=69.6μF, Lp=-36.4mH
1kHz: Z= 21.8Ω, Rs=21.3Ω, Cs=-35.2μF, Ls=719uH, Rp=22.3Ω, Cp=-1.51μF, Lp=16.7mH
10kHz: Z= 14.2Ω, Rs=14.2Ω, Cs=87.5μF, Ls=-2,90μH, Rp=14.2Ω, Cp=14.4nF, Lp=-17.6mH
50kHz: Z= 18.5Ω, Rs=15.7Ω, Cs=-326nF, Ls=31.1μH, Rp=21.7Ω, Cp=-91.2nF, Lp=111μH
100kHz: Z= 25.5Ω, Rs=18.5Ω, Cs=-90.4nF, Ls=28μH, Rp=35.2Ω, Cp=43.1nF, Lp=58.8μH
In the frequency range of interest, i.e., where we observe ripples in the frequency response, the inductive component of the impedance is less than 100 μH. For the series model, it is approximately 30 μH, and for the parallel model, we have an inductance of about 60 μH in parallel with a capacitance of approximately 40 nF. My meter is not capable of measuring impedance at higher frequencies, but there is a visible trend of a slight decrease in the series inductance component and a much more significant decrease for the parallel inductance.
This impedance meter is a "substitute" RLC bridge utilizing the Cyrustek ES168/ES169 chipset, whose description, along with the measurement method and calculation algorithms, is detailed here: http://www.cyrustek.com.tw/wp-content/uploads/ES168.pdf
Quite interesting results.
A slow rise of impedance towards higher frequencies can be seen.
At the LC resonant frequency the real part of impedance is too high for effective dampening -
and this is the reason for more resonant peaking compared to low inductance dummies load setup.
Anyway there is no perfect solution for any loudspeaker load.
But I agree that +-0,5dB frequency ripple inside 20~20kHz should be fine.
Your serious investigations are much appreciated.😎
A slow rise of impedance towards higher frequencies can be seen.
At the LC resonant frequency the real part of impedance is too high for effective dampening -
and this is the reason for more resonant peaking compared to low inductance dummies load setup.
Anyway there is no perfect solution for any loudspeaker load.
But I agree that +-0,5dB frequency ripple inside 20~20kHz should be fine.
Your serious investigations are much appreciated.😎
An attempt to transform the output snubbers into Zobel networks. My choice is a capacitance of 100nF, meaning the snubbers 3k3 + 10n become Zobel networks of 100nF. At the same time, 100nF MLCC capacitors with a C0G dielectric and a rated operating voltage of 50V or 100V (100V recommended) are available. However, it becomes necessary to change the inductance of the filter coils from 10μH to 7μH. The capacitance of C2 (anode to ground in the tube stage) should be increased to 390pF. Determination of carrier frequency attenuation, which is slightly lower at 39dB. Below are the simulation results for this modified circuit.
Frequency response characteristics:
Zobel network analysis for three amplifier load scenarios:
Determination of carrier frequency attenuation for three amplifier load scenarios:
Final schematic (for simulation):
I also performed a simulation for filters with 10μH + 150nF and with the capacitance of capacitor C2 equal to 360pF. The first graph shows the frequency response characteristics, while the second graph presents the Zobel network analysis.
Frequency response characteristics:
Zobel network analysis for three amplifier load scenarios:
Determination of carrier frequency attenuation for three amplifier load scenarios:
Final schematic (for simulation):
I also performed a simulation for filters with 10μH + 150nF and with the capacitance of capacitor C2 equal to 360pF. The first graph shows the frequency response characteristics, while the second graph presents the Zobel network analysis.
There is an error in the attached schematic: the resistors R7 and R8 have incorrect labels. Their correct value, used in the calculations, is 3k3. Here's the corrected schematic:
Lo is the element used to simulate the operation with a real load.
Lo is the element used to simulate the operation with a real load.
Thanks.
Contrary to TPA32xx the TPA31xx provides a gain setting option up to 36dB.
Interesting feature, combined with PFFB to improve overall linearity.
Actually I am fiddling with that approach and it looks promising
Contrary to TPA32xx the TPA31xx provides a gain setting option up to 36dB.
Interesting feature, combined with PFFB to improve overall linearity.
Actually I am fiddling with that approach and it looks promising
Attachments
Last edited:
The ability to set a higher gain in the TPA3126/TPA3156 chip allows for the use of a deeper PFF loop, which is undoubtedly tempting. However, it is important to keep in mind the risk of instability, especially since the actual, built circuit will differ from the one "drawn" for simulation. Additionally, the complexity of crossover networks in speaker systems complicates both the analysis and the entire design process. A fundamental step in improving the THD level is the use of better inductors with superior cores, even at the cost of larger dimensions and higher expenses. Of course, this won’t be a super-miniature amplifier, and for such purposes, it is necessary to reevaluate the design assumptions. In this thread, I am presenting the optimization of a prototype stationary hybrid amplifier with a tube-based preamp stage, so maximizing energy efficiency and minimizing size are not the primary design criteria. Higher costs need to be accepted. If this is an issue, then unfortunately, we have to lower the quality bar, meaning using lower-quality components. This implies we are approaching the level of cheap Aliexpress modules, which unfortunately suffer from design flaws, low-quality PCBs, and poor layout—but they are cheap. When choosing a DIY project, the decision boils down to whether we want something better than mass-produced options or something of similar quality to mass production, but self-assembled. It is also important to distinguish whether we are talking about a full design, a prototype, corrections, and the development of a final version, or simply assembling a ready-made kit. Thus, the hobbyist designer or amateur assembler must decide what they want. Personally, I aim to be closer to the first-mentioned options; I’m not a top electronics expert, just a hobbyist. This project is essentially for personal satisfaction, and if it is well-received by a few testers, that would make it even more rewarding. Another important conclusion from this discussion is that we must choose a compromise—there is no perfect solution. That concludes the introduction and non-technical matters.
I will also add measurements made with an impedance meter of the full-range speaker with a nominal impedance of 8Ω for the series connection of resistance and inductance:
100 Hz: 510μH, 7.84Ω
400 Hz: 211μH, 7.91Ω
1 kHz: 75.3μH, 8.36Ω
4 kHz: 123μH, 8.28Ω
10 kHz: 106.2μH, 9.66Ω
40 kHz: 71.4μH, 16.0Ω
100 kHz: 52.7μH, 25.9Ω
What can be observed is the fact that, unlike my measurement of a speaker system, there are no capacitive components or negative values for impedance or capacitance here, which would indicate that for the given test frequency, we are examining a reactive circuit that deviates significantly from the model of a series or parallel connection of resistance and reactance, which would be an inductor or capacitor. Nevertheless, for frequencies beyond the audible range, that is, in the range critical for optimizing the behavior of the analyzed amplifier, it can be assumed that the crossover in the loudspeaker can already be described as a series connection of resistance and inductance. The measurements suggest slightly lower values of the resistance and inductive reactance components in the case of a speaker system with a crossover. However, it should be noted that only one speaker was measured, albeit with a relatively conservative crossover design, but we may encounter more complex crossover systems. Additionally, the impedance characteristics are also influenced by speaker resonances, the enclosure, and so on. Only the real components of the current—meaning power—contribute to the work performed by the speakers (plus, of course, heat dissipation), while the reactive or passive components merely sustain electric and magnetic fields and cannot be converted into work (or heat).
Now, it’s time for further experimentation with the use of Zobel networks. First, here is the schematic diagram of the entire amplifier:
In the first step, high-pass filtering in the preamp stage is removed by setting the value of capacitor C2 to 1pF, as this better reflects what is happening with the power stage. Below are the simulation results for my chosen component values in the PFF loop and Zobel networks.
Amplitude-frequency characteristics, including an attempt to estimate carrier suppression of the switching signal:
Unfortunately, we have a slightly worse, although still decent, result—not worse than 39 dB for the worst of the analyzed combinations of values in the simulated circuit. This is slightly worse than the 44 dB level for the initial case discussed earlier this week.
The next point is the ringing for a square wave signal, including the case of no connected load (where the load becomes a 2k2 resistor), showing the entire square wave impulse and a magnified area of ringing with overshoots. The simulation results for ringing are considered at least good:
The next stage involves introducing a high-pass filter in the preamp stage, exactly as described in the publication by Texas Instruments. Below are the results (I am skipping the THD analysis, as these modifications do not significantly impact THD levels):
In summary, it is also evident that at 20 kHz, there will be slightly greater phase shift, but still below 25º for 8Ω loads. For the 16 Hz–16 kHz range, it fits within the range of +20º to -20º for various hypothetical amplifier loads. Of course, it should also be noted that we are optimizing the range beyond the acoustic band. While the case of no load is important, these additional resonances are not particularly significant. Similarly, the ringing is not something that substantially degrades the reproduced sound. We won’t be listening to square waves (unless someone is a masochist), and attempting to reproduce them, as well as high-frequency sine waves at full output power, will essentially damage the tweeters. In reproduced music, the levels of high frequencies are minimal, as shown in the example spectrogram.
In my opinion, further efforts to improve these details around 100 kHz likely do not make much sense - they would merely serve as training in using the TINA program or as an intellectual exercise. In this case, I prefer to focus on other issues.
Unfortunately, I also need to change the PCB design because ceramic capacitors of 220 nF at 100 V with a dielectric that does not exhibit strong capacitance dependence on voltage (e.g., C0G and U2J) are not readily available. I must use film capacitors - polypropylene capacitors capable of handling voltage rise rates of around 100 V/μs - instead of MLCC in the Zobel networks. Perhaps 50 V/μs would suffice, but polyester capacitors or those in smaller cases are hard to find. These are relatively large components.
Best regards.
I will also add measurements made with an impedance meter of the full-range speaker with a nominal impedance of 8Ω for the series connection of resistance and inductance:
100 Hz: 510μH, 7.84Ω
400 Hz: 211μH, 7.91Ω
1 kHz: 75.3μH, 8.36Ω
4 kHz: 123μH, 8.28Ω
10 kHz: 106.2μH, 9.66Ω
40 kHz: 71.4μH, 16.0Ω
100 kHz: 52.7μH, 25.9Ω
What can be observed is the fact that, unlike my measurement of a speaker system, there are no capacitive components or negative values for impedance or capacitance here, which would indicate that for the given test frequency, we are examining a reactive circuit that deviates significantly from the model of a series or parallel connection of resistance and reactance, which would be an inductor or capacitor. Nevertheless, for frequencies beyond the audible range, that is, in the range critical for optimizing the behavior of the analyzed amplifier, it can be assumed that the crossover in the loudspeaker can already be described as a series connection of resistance and inductance. The measurements suggest slightly lower values of the resistance and inductive reactance components in the case of a speaker system with a crossover. However, it should be noted that only one speaker was measured, albeit with a relatively conservative crossover design, but we may encounter more complex crossover systems. Additionally, the impedance characteristics are also influenced by speaker resonances, the enclosure, and so on. Only the real components of the current—meaning power—contribute to the work performed by the speakers (plus, of course, heat dissipation), while the reactive or passive components merely sustain electric and magnetic fields and cannot be converted into work (or heat).
Now, it’s time for further experimentation with the use of Zobel networks. First, here is the schematic diagram of the entire amplifier:
In the first step, high-pass filtering in the preamp stage is removed by setting the value of capacitor C2 to 1pF, as this better reflects what is happening with the power stage. Below are the simulation results for my chosen component values in the PFF loop and Zobel networks.
Amplitude-frequency characteristics, including an attempt to estimate carrier suppression of the switching signal:
Unfortunately, we have a slightly worse, although still decent, result—not worse than 39 dB for the worst of the analyzed combinations of values in the simulated circuit. This is slightly worse than the 44 dB level for the initial case discussed earlier this week.
The next point is the ringing for a square wave signal, including the case of no connected load (where the load becomes a 2k2 resistor), showing the entire square wave impulse and a magnified area of ringing with overshoots. The simulation results for ringing are considered at least good:
The next stage involves introducing a high-pass filter in the preamp stage, exactly as described in the publication by Texas Instruments. Below are the results (I am skipping the THD analysis, as these modifications do not significantly impact THD levels):
In summary, it is also evident that at 20 kHz, there will be slightly greater phase shift, but still below 25º for 8Ω loads. For the 16 Hz–16 kHz range, it fits within the range of +20º to -20º for various hypothetical amplifier loads. Of course, it should also be noted that we are optimizing the range beyond the acoustic band. While the case of no load is important, these additional resonances are not particularly significant. Similarly, the ringing is not something that substantially degrades the reproduced sound. We won’t be listening to square waves (unless someone is a masochist), and attempting to reproduce them, as well as high-frequency sine waves at full output power, will essentially damage the tweeters. In reproduced music, the levels of high frequencies are minimal, as shown in the example spectrogram.
In my opinion, further efforts to improve these details around 100 kHz likely do not make much sense - they would merely serve as training in using the TINA program or as an intellectual exercise. In this case, I prefer to focus on other issues.
Unfortunately, I also need to change the PCB design because ceramic capacitors of 220 nF at 100 V with a dielectric that does not exhibit strong capacitance dependence on voltage (e.g., C0G and U2J) are not readily available. I must use film capacitors - polypropylene capacitors capable of handling voltage rise rates of around 100 V/μs - instead of MLCC in the Zobel networks. Perhaps 50 V/μs would suffice, but polyester capacitors or those in smaller cases are hard to find. These are relatively large components.
Best regards.
Once again, I missed updating the labels on the schematic. It concerns the resistors in the Zobel networks, which are labeled as 3R3 on the schematic, but their actual values were 2.2Ω. The waveforms for ringing were performed with the capacitor C2 set to 390pF.
Out of curiosity, I also ran simulations with these resistors in the Zobel circuits set to 3.3Ω (C2 = 560pF). This small change is beneficial, as shown below:
The suppression of the switching carrier is 39dB.
Out of curiosity, I also ran simulations with these resistors in the Zobel circuits set to 3.3Ω (C2 = 560pF). This small change is beneficial, as shown below:
The suppression of the switching carrier is 39dB.
Agreed in all aspects - except the one using tubes.
My last tube build is >50yrs ago.
But that's me.
My last tube build is >50yrs ago.
But that's me.
The use of vacuum tubes is, after all, a matter of sentiment. I grew up in a communist-era apartment block, and electronics in the USSR lagged behind. After the fall of the USSR, many military components eventually made their way into civilian circulation, including interesting rod tubes uniquely produced in the USSR. Their unique electrode design provided them with rather specific properties, such as zero transit time between the cathode plane and the control grid. During the communist era, I could only read about subminiature tubes or nuvistors, although American-made nuvistors occasionally appeared in the market as surplus stock from civilian state-owned companies. However, their price was virtually prohibitive. Another interesting tidbit was the plan to use radioactive elements as vacuum tube cathodes. Fortunately, this idea was abandoned in favor of specific field-emission cathodes made from magnesium. Interestingly, this concept has recently resurfaced with the idea of using carbon nanotubes. It's also worth mentioning the use of micro-triodes in integrated circuits; several years ago, this topic was the subject of numerous scientific publications.
I mentioned earlier that Soviet nuvistors differ somewhat from the originals, so I’m presenting a photo of a Soviet nuvistor. As for rod tubes, I refer you to -> https://www.radiomuseum.org/forum/russian_subminiature_tubes.html
I mentioned earlier that Soviet nuvistors differ somewhat from the originals, so I’m presenting a photo of a Soviet nuvistor. As for rod tubes, I refer you to -> https://www.radiomuseum.org/forum/russian_subminiature_tubes.html
Something more devilish is swirling in my head. To combine the idea of using a triode as a nonlinear element in a negative feedback loop (i.e., the concept of super triode connected amplifier) with a PFF loop.
I personally love the sound of 6N1P tubes as a SET buffer. They are very robust. I use that on my hybrid opamp / tube buffer and thinking of using this on my TPA3255 with an OPA1637 balanced line driver from the tube to the TPA3255.
@STUDI - what is your recommendation summary for mods to TI’s recommended PFFB for a TPA3255 driving 8ohm loads in BTL stereo mode?
@STUDI - what is your recommendation summary for mods to TI’s recommended PFFB for a TPA3255 driving 8ohm loads in BTL stereo mode?
Last edited:
This is a good tube (often mistakenly treated as a substitute for the ECC88; only the gain factor is nearly the same, but all other electrical parameters are different). A drawback of this tube is its significant filament current. Unlike the ECC88, utilizing its potential requires a substantial anode voltage. For a project like the one I’m presenting here, it’s not a good choice. I recommend the less power-hungry Soviet-Union duotriodes 6N3P-E / 6N3P-EV (with an unusual filament pin layout – 1 and 9 instead of 4 and 5) – a strict equivalent to the 2C51, 5670, 6CC40 tubes.
Firstly, using 8-ohm speakers requires a different combination of inductance and capacitance for the output filters, although the starting point will still depend on the available inductors. Good options seem to be the double, albeit non-coupled, UA8013-AL and UA8014-AL. Slightly higher gain will allow achieving a slightly deeper PFF loop. As a starting point, the values from Texas Instruments’ publications should be considered.
However, simulations with standard circuit components will unfortunately diverge significantly from the actual circuit performance. Texas Instruments does not provide a Spice model for this family of chips, which leaves designing a test PCB for experiments as the only option. This, unfortunately, comes with additional costs.
Even using tools like the QA-403 and QA-451 (QuantAsylum), it’s not possible to obtain an amplitude-frequency response curve above 40–50 kHz, which is critical for assessing stability, ringing, and the full bandwidth. A 7th-order filter with a cutoff frequency of around 67 kHz won’t allow this but will enable THD measurements at base frequencies higher than a few kHz. At least the 2nd and 3rd harmonics will be captured in the measurement for a test frequency of 22 kHz.
This sharp filtering above 24 kHz, for which the parameters of Class D amplifier chips are typically specified, unfortunately does not allow for a reliable assessment of THD at frequencies above 4 kHz. A filter added to eliminate switching residuals, with a bandwidth of 80 kHz and a lower order, will attenuate the amplitude of both the signal and its harmonics starting from around 20–30 kHz.
This approach doesn’t convince me. While it is often argued that we "shouldn't hear" these harmonics, this method "accidentally" improves the reported THD values in datasheets.
Firstly, using 8-ohm speakers requires a different combination of inductance and capacitance for the output filters, although the starting point will still depend on the available inductors. Good options seem to be the double, albeit non-coupled, UA8013-AL and UA8014-AL. Slightly higher gain will allow achieving a slightly deeper PFF loop. As a starting point, the values from Texas Instruments’ publications should be considered.
However, simulations with standard circuit components will unfortunately diverge significantly from the actual circuit performance. Texas Instruments does not provide a Spice model for this family of chips, which leaves designing a test PCB for experiments as the only option. This, unfortunately, comes with additional costs.
Even using tools like the QA-403 and QA-451 (QuantAsylum), it’s not possible to obtain an amplitude-frequency response curve above 40–50 kHz, which is critical for assessing stability, ringing, and the full bandwidth. A 7th-order filter with a cutoff frequency of around 67 kHz won’t allow this but will enable THD measurements at base frequencies higher than a few kHz. At least the 2nd and 3rd harmonics will be captured in the measurement for a test frequency of 22 kHz.
This sharp filtering above 24 kHz, for which the parameters of Class D amplifier chips are typically specified, unfortunately does not allow for a reliable assessment of THD at frequencies above 4 kHz. A filter added to eliminate switching residuals, with a bandwidth of 80 kHz and a lower order, will attenuate the amplitude of both the signal and its harmonics starting from around 20–30 kHz.
This approach doesn’t convince me. While it is often argued that we "shouldn't hear" these harmonics, this method "accidentally" improves the reported THD values in datasheets.
Last edited:
I’m using it at 90v as a drop in substitute for a design nominally based on 6JD6 or E88CC. But it measures well and sounds great. It does draw a lot of current but I am using a DCDC converter with good efficiency to generate the 90v from 12v input. The tube output is buffered with OPA1637 driver.
Here is circuit driving 2k2 load at 2.83vrms and getting about 0.062% THD.
Swapping to E88CC we get 0.00042% THD. Sound is very close to sound of a solid state buffer.
Here is circuit driving 2k2 load at 2.83vrms and getting about 0.062% THD.
Swapping to E88CC we get 0.00042% THD. Sound is very close to sound of a solid state buffer.
Two Audible Hum Sources
Analyzing and Improving THD Spectrum Performance
- 60 Hz Peak – AC Hum Leakage:
- The primary source of this peak is the AC hum penetration from tube filament heating. This issue arises from unavoidable leakage currents in mains-powered devices and suboptimal shielding. One aspect worth examining is the leakage between the filament and the cathode in the tube. This leakage intensifies over time due to electrolytic degradation of the insulating material (typically alumina).
- As per datasheets of high-reliability tubes, it is required that the resistance between the cathode and the filament should not exceed 20 kΩ. Ignoring this can lead to worsening performance over time.
- Another overlooked issue is the parasitic diode formed by the exposed, heated filament ends and other electrodes. A recommended solution is to elevate the filament voltage relative to the circuit ground. While this doesn’t accelerate the degradation of the cathode-filament insulation, it significantly reduces parasitic diode currents.
- Telefunken demonstrated this in the 1960s in their Laborbuch, using an ECC808 tube in a preamplifier. They compared two designs: one with automatic cathode bias (RC network between the cathode and ground) and another with grid current polarization (a high-value grid resistor of 10–22 MΩ). The latter design showed significantly lower noise levels.
- Full-Wave Rectification Residue – Ripple from Anode Voltage:
- The second hum peak comes from ripple in the anode voltage, a byproduct of full-wave rectification. Mains power typically contains odd harmonics of the fundamental frequency. Core saturation in transformers and the inherent nonlinearity of rectification (similar to switching circuits) generally prevent even harmonics from being generated.
- This hum peak suggests a problem with anode voltage filtration and calls for improvements to the PCB layout. Adopting a star-ground topology can help eliminate ground loops and reduce noise.
- Another contributing factor could be the use of standard rectifier diodes or bridges. These components, rooted in 1960s technology, generate noise due to surge currents at zero crossings of the mains voltage. Carrier dynamics within the semiconductor material don’t cease instantaneously when the electric field disappears, causing noise.
- Upgrading to Fast Recovery diodes can mitigate this issue, significantly reducing noise from surge currents. Additionally, while bypassing rectifier diodes with small capacitors (2–22 nF) is a common practice to suppress high-frequency noise, it’s only a partial solution. Modern power systems often exhibit numerous harmonics in the mains voltage, with higher-order harmonics (e.g., 21st and above) becoming increasingly prevalent. High-quality power analyzers now measure up to at least the 51st harmonic of 50/60 Hz to provide a comprehensive view of power quality.
Analyzing and Improving THD Spectrum Performance
- Observations from the THD Spectrum:
- The 3rd harmonic being at the same level as the 2nd harmonic indicates suboptimal operating conditions for the tube. This is a result of the nonlinear plate characteristics at low anode voltages and currents.
- Excessive signal amplitudes at the grid exacerbate these nonlinearities, pushing the tube beyond its linear operating region.
- The second measurement suggests the tube is far from full drive, meaning there is room for improvement by adjusting operating parameters.
- Recommendations for Improvement:
- Increase the Plate Supply Voltage:
- The 6N1P tube performs better at higher anode voltages. Increasing the plate voltage from 90V to around 250V is recommended. This places the tube in a more linear region of its characteristic curves, reducing distortion.
- Adjust Signal Levels:
- Ensure that the input signal levels at the grid remain within a range that prevents excessive swing. This avoids driving the tube into grid conduction or saturation, which are major sources of higher-order harmonics.
- Optimize Harmonic Balance:
- As a general guideline, aim for the 3rd harmonic to be at least 20 dB lower than the 2nd harmonic at a -3 dB output level. This represents a one-order-of-magnitude reduction and ensures smoother, more pleasant sound reproduction.
- The 2nd harmonic can remain at a relatively high level (up to 1%, or -40 dB) without being perceived as distortion, as the human ear tolerates octave-based harmonic content. Psychoacoustically, it blends into the original signal, often perceived as warmth or richness.
- Increase the Plate Supply Voltage:
- Psychoacoustic Insights:
- The human auditory system is less sensitive to harmonic distortion that aligns with musical intervals, such as octaves (e.g., 2nd harmonic). This is due to the way our ears and brain process harmonic relationships.
- However, the 3rd harmonic and higher odd harmonics are more likely to be perceived as harshness or distortion, especially if they approach or exceed the level of the 2nd harmonic. This makes their suppression critical.
Last edited:
Thanks for the detailed response. This is a TPA3255 thread so I dont want to clog it up with tube stuff. I totally agree with you about harmonic profile and psycho acoustics of 3Rd orders and odd orders being lower than 2nd for most pleasing sound. I usually design harmonic profile of my amps to have about -10dB lower 3rd and monotonic descending higher orders. Getting 250v is hard on this particular design as DCDC converter was designed for 90v for E88CC. The 60Hz hum actually is not audible at -100dB. It’s probably from EMI pickup va filament because filament heater is from 60kHz DCDC switch mode converter.
The values of these peaks are small. Both the equal-loudness curves, the characteristics of loudspeaker systems, and the background acoustic noise will "help" make the 60Hz (50Hz) hum unnoticeable. Achieving this effect is more difficult for 120Hz (100Hz). On the other hand, I can envy the truly clean power supply from the electrical grid. In my conditions, this is unattainable (inverters for renewable energy and the dominance of switching power supplies unfortunately contribute significant harmonics and subharmonics).
There is also another source of interference. I assume a computer/laptop was used during the measurements. Switching power supplies often have an EMI filter on the input, frequently with so-called artificial ground if the power connection lacks a protective earth conductor. The capacitance between the artificial ground point (a capacitive divider between the neutral and live wires) and the power supply output does not exceed 2.5nF (for safety reasons, 2.3nF at 230V AC corresponds to an impedance of 1.38 MΩ, resulting in a current of approximately 0.2 mA, which statistically exceeds the perception threshold for women). Even if no such additional capacitance is present, parasitic capacitances will still exist between the windings of the transformer (both switching and traditional) and from mounting arrangements.
The switching power supply, when powered from the mains, can generate hum with numerous harmonics. Additionally, a DC/DC converter for tube filament heating can inject a 120Hz hum into the system.
Yes, I know we are straying from the main topic, but since this digression has already arisen and these are also important issues, it is worth mentioning them here. Addressing these matters in other threads or forums would, in practice, lead to them being overlooked.
As for the argument that a DC/DC converter supplying 90V already exists—I agree with this point. However, if we want to use tubes other than those similar to the ECC88, which are designed for cascode RF stages in VHF tuners, and instead choose tubes whose characteristics require a more typical plate voltage closer to 250V, this will mean designing a new converter and replacing the capacitors in the tube circuit with ones rated for higher voltages.
The effectiveness of ripple filtering or power decoupling depends on the ESR of the capacitor. For electrolytics, we face a problem: high-quality organic polymer electrolytic capacitors with low ESR are not manufactured for such high voltages. Their typical range is between 63V and 100V. Additionally, these polymer capacitors exhibit excellent ESR and impedance characteristics as a function of frequency, which is especially desirable when using DC/DC converters as power sources.
Another issue is the behavior at the output of the stage when the plate voltage is applied, removed, or during the warm-up of the heater as electron emission begins. The RC coupling circuit at the output acts as a differentiator. When the plate voltage is turned on, a positive high-voltage pulse occurs, and when it is turned off, a negative pulse follows. The peak value and duration of these pulses depend on the rise and fall time of the plate power supply voltage.
The situation is worse with the onset of cathode emission, as the pulse may have a relatively low peak value but can last quite a long time, introducing interference into the output signal fed to subsequent stages. These impulse surges are also a hazard to the next stages and require protective measures.
Typical protective measures include:
I have seen cases where a tube preamplifier damaged the semiconductor power stage connected to its output (this involved products from the high-end audiophile segment, where additional protective elements are often considered "bad" in the design philosophy).
Finally, on the subject of capacitance between the windings of a mains transformer: someone once asked why a neon tester lights up when brought near any unconnected terminal of its secondary winding, even though measurements confirm excellent insulation between the windings.
I had a 50W mains transformer at hand and measured the impedance between its windings with an impedance meter. The capacitance was about 14pF, which at 50Hz corresponds to a reactance of approximately 230MΩ, resulting in a leakage current of about 1µA. This is enough to light small neon bulbs.
By writing about this in this thread, I am, of course, not implying any lack of knowledge on your part. However, since this is a DIY forum, I believe it’s important not to avoid these topics, as others attempting to replicate hybrid designs using vacuum tubes may unknowingly cause damage or face failures in their projects.
This has turned into quite a bit of off-topic discussion, but it remains a significant portion of essential knowledge.
Best regards.
There is also another source of interference. I assume a computer/laptop was used during the measurements. Switching power supplies often have an EMI filter on the input, frequently with so-called artificial ground if the power connection lacks a protective earth conductor. The capacitance between the artificial ground point (a capacitive divider between the neutral and live wires) and the power supply output does not exceed 2.5nF (for safety reasons, 2.3nF at 230V AC corresponds to an impedance of 1.38 MΩ, resulting in a current of approximately 0.2 mA, which statistically exceeds the perception threshold for women). Even if no such additional capacitance is present, parasitic capacitances will still exist between the windings of the transformer (both switching and traditional) and from mounting arrangements.
The switching power supply, when powered from the mains, can generate hum with numerous harmonics. Additionally, a DC/DC converter for tube filament heating can inject a 120Hz hum into the system.
Yes, I know we are straying from the main topic, but since this digression has already arisen and these are also important issues, it is worth mentioning them here. Addressing these matters in other threads or forums would, in practice, lead to them being overlooked.
As for the argument that a DC/DC converter supplying 90V already exists—I agree with this point. However, if we want to use tubes other than those similar to the ECC88, which are designed for cascode RF stages in VHF tuners, and instead choose tubes whose characteristics require a more typical plate voltage closer to 250V, this will mean designing a new converter and replacing the capacitors in the tube circuit with ones rated for higher voltages.
The effectiveness of ripple filtering or power decoupling depends on the ESR of the capacitor. For electrolytics, we face a problem: high-quality organic polymer electrolytic capacitors with low ESR are not manufactured for such high voltages. Their typical range is between 63V and 100V. Additionally, these polymer capacitors exhibit excellent ESR and impedance characteristics as a function of frequency, which is especially desirable when using DC/DC converters as power sources.
Another issue is the behavior at the output of the stage when the plate voltage is applied, removed, or during the warm-up of the heater as electron emission begins. The RC coupling circuit at the output acts as a differentiator. When the plate voltage is turned on, a positive high-voltage pulse occurs, and when it is turned off, a negative pulse follows. The peak value and duration of these pulses depend on the rise and fall time of the plate power supply voltage.
The situation is worse with the onset of cathode emission, as the pulse may have a relatively low peak value but can last quite a long time, introducing interference into the output signal fed to subsequent stages. These impulse surges are also a hazard to the next stages and require protective measures.
Typical protective measures include:
- Clamping Diodes: One reverse-biased to ground or the negative supply and another to the positive supply of the stage.
- Zener Diodes: Using a Zener diode with a low reverse current. Contrary to some opinions, this will not degrade the sound quality (although the diode's capacitance could occasionally pose a problem, which must be analyzed).
I have seen cases where a tube preamplifier damaged the semiconductor power stage connected to its output (this involved products from the high-end audiophile segment, where additional protective elements are often considered "bad" in the design philosophy).
Finally, on the subject of capacitance between the windings of a mains transformer: someone once asked why a neon tester lights up when brought near any unconnected terminal of its secondary winding, even though measurements confirm excellent insulation between the windings.
I had a 50W mains transformer at hand and measured the impedance between its windings with an impedance meter. The capacitance was about 14pF, which at 50Hz corresponds to a reactance of approximately 230MΩ, resulting in a leakage current of about 1µA. This is enough to light small neon bulbs.
By writing about this in this thread, I am, of course, not implying any lack of knowledge on your part. However, since this is a DIY forum, I believe it’s important not to avoid these topics, as others attempting to replicate hybrid designs using vacuum tubes may unknowingly cause damage or face failures in their projects.
This has turned into quite a bit of off-topic discussion, but it remains a significant portion of essential knowledge.
Best regards.
Last edited:
What I can also commend is the absence of visible 1/f noise. It’s likely below the resolution of the ADC converter.
1/f noise is, unfortunately, present in electronic circuits (including the measurement equipment used), such as thermal noise. However, tubes introduce shot noise, which accompanies the random nature of electron emission from the cathode. Since this noise predominantly affects low frequencies (below 1kHz), even a relatively high level at 10Hz will only slightly degrade the overall SNR. On the other hand, this 1/f noise appears as a "grass" baseline in the FFT when we switch from a logarithmic to a linear frequency scale.
1/f noise is, unfortunately, present in electronic circuits (including the measurement equipment used), such as thermal noise. However, tubes introduce shot noise, which accompanies the random nature of electron emission from the cathode. Since this noise predominantly affects low frequencies (below 1kHz), even a relatively high level at 10Hz will only slightly degrade the overall SNR. On the other hand, this 1/f noise appears as a "grass" baseline in the FFT when we switch from a logarithmic to a linear frequency scale.