The influence of high frequencies (radio frequencies and beyond) on electronic circuits is sometimes put forward as a possible cause to explain audible effects or subjective assessments of the sound quality of a specific device in an audio chain. Notwithstanding the difficulties in establishing and confirming the reality of what is claimed to be heard, especially when the audible effects are presented as being subtle or difficult to describe, or when they are evanescent or circumstantial, I have yet read only anecdotal evidence on the effects of supposed high frequencies or hypotheses on the mechanism by which these high frequencies could influence the irrevocably audible frequency band (20 Hz to 20 kHz).
However, a few years ago I was able to note the appearance, in some application notes and datasheets recently published by manufacturers of integrated circuits, of a relatively new specification concerning integrated operational amplifiers (op amp): the rate of EMIRR (for EMI Rejection Ratio), which is therefore a rejection rate of common modes (the signals between each of the inputs of an op amp and the ground) specifically measured in the high frequency domain. This specification is applicable to some integrated op amps which are sensitive to radio-frequency interference: they see their offset voltage vary depending on the level and frequency spectrum of this interference. Manufacturers of integrated circuits have therefore been able to identify an effect of high frequencies which modifies an operating parameter of their integrated op amps (not all are equal in the face of this phenomenon). That being said, I only mention this subject to illustrate the fact that the influence of high frequencies is neither a myth nor a simple hypothesis. I am not claiming that the phenomenon assessed by this particular specification necessarily matters in audio.
But I finally came across several documents published by Japanese authors which clearly highlighted the influence of high frequencies (from the order of megahertz to hundreds of megahertz) in the audio band analyzed at the output of different power amplifiers. The phenomenon described in these documents is in itself nothing new: it involves intermodulation distortion measurements in the presence of a composite signal of two close frequencies of the same level.
If the phenomenon described by the Japanese is nothing new, what is, to my knowledge, unprecedented is the realization and especially the publication of intermodulation measurements in the audio band with very high frequency excitation signals. These experiment is therefore worth a discussion.
I will rely on one of the documents that I mentioned: The influence of non-audible plural high frequency electrical noise on the playback sound of audio equipment (2nd report), by N. Kimura and T . Yoshida (reference: IOP Conf. Series: Journal of Physics: Conf. Series 1234567890 1075 (2018) 012006; DOI: 10.1088/1742-6596/1075/1/012006). This document being published under the Creative Commons license, I am attaching this document below, without alteration or modification.
Here are illustrations taken from this document.
The experimental setup used by the authors is as follows:
There is nothing special to report: an arbitrary function generator is programmed to produce two sinusoidal frequencies which are mixed to create a composite signal injected directly into the input of the device under test (shown in salmon color), of which the output is observed using a well-known Audio Precision APx525 analyzer. The two frequencies are chosen such that second-order intermodulation produces a frequency of 2,017 hertz (Hz), right in the midrange in the audio band. The 2nd order intermodulation is the frequency which corresponds to the difference (in hertz) of the two frequencies of the test signal. The attenuator before the device under test is only used to make measurements at different excitation signal levels. The gains of the different amplifiers are aligned to obtain the same output level of 0 dBV (1 V RMS) on a sine of 2,017 Hz.
The authors tested six amplifiers of different brands and models: three linear amplifiers and three switching amplifiers (i.e. class D amplifiers). The brands and references of the amplifiers have not been disclosed by the authors. They carried out measurements on each device by varying the first frequency of the composite signal from 0.1 MHz (100 kHz) to 100 MHz. They also carried out measurements with a composite signal whose amplitude was varied from -35 dBV (17.8 mV RMS) to 0 dBV (1 V RMS).
Here are excerpts from the measurement results.
This graph shows the level of 2nd order distortion (the frequency of 2,017 Hz) measured at the output of each amplifier as a function of the first frequency of the composite signal for a test signal level of 0 dBV:
A, B and C are the linear amplifiers; D, E and F the class D amplifiers. We can see that the behavior of each device is very different and independent of the class of amplification. We can also see that with an excitation signal of 1 V RMS, some amplifiers produce, at some frequencies of the excitation signal, a 2nd order distortion greater than -40 dB (1%) or even -20 dB (10%).
But 1 V RMS is a rather very high signal amplitude for high frequencies. The authors therefore also measured the amplifiers at different excitation signal levels. Here is a graph which summarizes these results for the three worst amplifiers according to the previous measurement, the A and the C (linear amplifiers) and the E (class D amplifier; this is the previous E device and not the F, which is clearly a typographical error), each of them being measured with a composite signal whose first frequency corresponds to the worst case previously observed:
The dotted curves are the average curves deduced from the measurement points. We can see with relief that the intermodulation distortion decreases with the level of the high frequency signal. To get the idea, -20 dBV corresponds to 100 mV RMS.
What preliminary lessons can we learn from these results? In fact, in my opinion, they open up more questions than answers. The signal injection mode (directly at the amplifier input) is not normally representative of a real practical case (subject to what follows in the penultimate paragraph). But with high frequencies, you have to be careful: the point of injection for parasitic signals are sometimes unexpected, especially as you go up in frequency. The levels of the excitation signals still seem very high to me. That being said, there is clearly a phenomenon to be observed and, possibly, to take into account, and one may wonder whether specification or measurements of high frequency susceptibility would not be useful information for assessing the performance of an amplifier. Indeed, all amplifiers are clearly not equal when it comes to high frequencies, regardless of the amplification technique used. Finally, these measurements were carried out with a relatively simple high frequency excitation signal. What about more complex, variable, or broad-spectrum signals?
One last word. If, obviously, all amplifiers do not have the same susceptibility to high frequencies, this is undoubtedly due to their design or their manufacturing quality, or perhaps simply to the adherence or not by the designer or manufacturer to good practices. In general, it is recommended to limit the bandwidth of an amplifier directly at the input to avoid, or at least mitigate, possible problems brought by high frequencies. Was this the case for the amplifiers that were tested? More generally, the problem in the audiophile world is that many designs put on the market deviate from good or common practices for obscure and sometimes irrational reasons. This isn't just true for amplifiers. If the level of high frequency signals which was used for the test seems very high and, at first analysis, not typical of what one may encounter in the real world, we must not forget that some audiophile sources sometimes have unusual characteristics which approach these levels. Without naming a brand, we can for example wonder what the amplifier C above would produce with a non-oversampling filterless DAC or a SA-CD/DSD DAC with inadequate low-pass filtering connected to it.
Isn't it time to investigate in more detail the performance of analog devices when they are subjected, voluntarily or involuntarily, to high frequency signals to remove any doubt about their behavior?
However, a few years ago I was able to note the appearance, in some application notes and datasheets recently published by manufacturers of integrated circuits, of a relatively new specification concerning integrated operational amplifiers (op amp): the rate of EMIRR (for EMI Rejection Ratio), which is therefore a rejection rate of common modes (the signals between each of the inputs of an op amp and the ground) specifically measured in the high frequency domain. This specification is applicable to some integrated op amps which are sensitive to radio-frequency interference: they see their offset voltage vary depending on the level and frequency spectrum of this interference. Manufacturers of integrated circuits have therefore been able to identify an effect of high frequencies which modifies an operating parameter of their integrated op amps (not all are equal in the face of this phenomenon). That being said, I only mention this subject to illustrate the fact that the influence of high frequencies is neither a myth nor a simple hypothesis. I am not claiming that the phenomenon assessed by this particular specification necessarily matters in audio.
But I finally came across several documents published by Japanese authors which clearly highlighted the influence of high frequencies (from the order of megahertz to hundreds of megahertz) in the audio band analyzed at the output of different power amplifiers. The phenomenon described in these documents is in itself nothing new: it involves intermodulation distortion measurements in the presence of a composite signal of two close frequencies of the same level.
If the phenomenon described by the Japanese is nothing new, what is, to my knowledge, unprecedented is the realization and especially the publication of intermodulation measurements in the audio band with very high frequency excitation signals. These experiment is therefore worth a discussion.
I will rely on one of the documents that I mentioned: The influence of non-audible plural high frequency electrical noise on the playback sound of audio equipment (2nd report), by N. Kimura and T . Yoshida (reference: IOP Conf. Series: Journal of Physics: Conf. Series 1234567890 1075 (2018) 012006; DOI: 10.1088/1742-6596/1075/1/012006). This document being published under the Creative Commons license, I am attaching this document below, without alteration or modification.
Here are illustrations taken from this document.
The experimental setup used by the authors is as follows:
There is nothing special to report: an arbitrary function generator is programmed to produce two sinusoidal frequencies which are mixed to create a composite signal injected directly into the input of the device under test (shown in salmon color), of which the output is observed using a well-known Audio Precision APx525 analyzer. The two frequencies are chosen such that second-order intermodulation produces a frequency of 2,017 hertz (Hz), right in the midrange in the audio band. The 2nd order intermodulation is the frequency which corresponds to the difference (in hertz) of the two frequencies of the test signal. The attenuator before the device under test is only used to make measurements at different excitation signal levels. The gains of the different amplifiers are aligned to obtain the same output level of 0 dBV (1 V RMS) on a sine of 2,017 Hz.
The authors tested six amplifiers of different brands and models: three linear amplifiers and three switching amplifiers (i.e. class D amplifiers). The brands and references of the amplifiers have not been disclosed by the authors. They carried out measurements on each device by varying the first frequency of the composite signal from 0.1 MHz (100 kHz) to 100 MHz. They also carried out measurements with a composite signal whose amplitude was varied from -35 dBV (17.8 mV RMS) to 0 dBV (1 V RMS).
Here are excerpts from the measurement results.
This graph shows the level of 2nd order distortion (the frequency of 2,017 Hz) measured at the output of each amplifier as a function of the first frequency of the composite signal for a test signal level of 0 dBV:
A, B and C are the linear amplifiers; D, E and F the class D amplifiers. We can see that the behavior of each device is very different and independent of the class of amplification. We can also see that with an excitation signal of 1 V RMS, some amplifiers produce, at some frequencies of the excitation signal, a 2nd order distortion greater than -40 dB (1%) or even -20 dB (10%).
But 1 V RMS is a rather very high signal amplitude for high frequencies. The authors therefore also measured the amplifiers at different excitation signal levels. Here is a graph which summarizes these results for the three worst amplifiers according to the previous measurement, the A and the C (linear amplifiers) and the E (class D amplifier; this is the previous E device and not the F, which is clearly a typographical error), each of them being measured with a composite signal whose first frequency corresponds to the worst case previously observed:
The dotted curves are the average curves deduced from the measurement points. We can see with relief that the intermodulation distortion decreases with the level of the high frequency signal. To get the idea, -20 dBV corresponds to 100 mV RMS.
What preliminary lessons can we learn from these results? In fact, in my opinion, they open up more questions than answers. The signal injection mode (directly at the amplifier input) is not normally representative of a real practical case (subject to what follows in the penultimate paragraph). But with high frequencies, you have to be careful: the point of injection for parasitic signals are sometimes unexpected, especially as you go up in frequency. The levels of the excitation signals still seem very high to me. That being said, there is clearly a phenomenon to be observed and, possibly, to take into account, and one may wonder whether specification or measurements of high frequency susceptibility would not be useful information for assessing the performance of an amplifier. Indeed, all amplifiers are clearly not equal when it comes to high frequencies, regardless of the amplification technique used. Finally, these measurements were carried out with a relatively simple high frequency excitation signal. What about more complex, variable, or broad-spectrum signals?
One last word. If, obviously, all amplifiers do not have the same susceptibility to high frequencies, this is undoubtedly due to their design or their manufacturing quality, or perhaps simply to the adherence or not by the designer or manufacturer to good practices. In general, it is recommended to limit the bandwidth of an amplifier directly at the input to avoid, or at least mitigate, possible problems brought by high frequencies. Was this the case for the amplifiers that were tested? More generally, the problem in the audiophile world is that many designs put on the market deviate from good or common practices for obscure and sometimes irrational reasons. This isn't just true for amplifiers. If the level of high frequency signals which was used for the test seems very high and, at first analysis, not typical of what one may encounter in the real world, we must not forget that some audiophile sources sometimes have unusual characteristics which approach these levels. Without naming a brand, we can for example wonder what the amplifier C above would produce with a non-oversampling filterless DAC or a SA-CD/DSD DAC with inadequate low-pass filtering connected to it.
Isn't it time to investigate in more detail the performance of analog devices when they are subjected, voluntarily or involuntarily, to high frequency signals to remove any doubt about their behavior?
Last edited:
