On reducing the switching distortions of transistor UMZCH classes B-AB

Practice shows that the search for the reasons for the “transistor” sound only in the field of OOS and the initial frequency response of the UMZCH gives results that are not always comparable to the efforts expended on the implementation of the complicated circuitry of the UMZCH and on the selection of high-quality components. Any industrial / home-made amplifier has “its own subtle sound signature”, which is expressed in the best (through it) sounding of phonograms of certain genres and music performers.

Ultimately, the amplifier plays an important role in shaping the taste, preferences of the owner (music lover), as well as the formation of his music library.

One of the types of distortions inherent in transistor push-pull amplifiers of B-AB classes are switching (commutation) distortions of the output stage transistors, when the active amplifying arm changes when the upper / lower half-wave of the actually reproduced music signal is reproduced.

At the moments of changing the active arms of the output stage of the UMZCH (subsets of p / elements on the physical path of the signal), such parameters as the output impedance (load damping factor), the OOS depth / initial gain, phase, etc. change abruptly.

This type of distortion as a class is absent in the UMZCH with a single-ended output stage. The most unpleasant thing is that the switching distortions are caused by nothing more than the amplified signal and their mechanism of influence on the quality of sound reproduction “de facto” has not yet been analyzed from the point of view of quantitative assessments.

Mathematical analysis of digital music phonograms, carried out using high-quality audio CDs, showed that in most cases, the polarity of the signal at the output of the UMZCH occurs from 2500 to 6000 times in 1 second (the dependence on the spectral composition of the musical phonogram is small).

Exactly so many times in the UMZCH there are prerequisites for the influence of switching distortions on its sound, and it is not known to what extent. it depends on the amplitude-time vector of the output voltage change (relative to the speed characteristics of the output stage) at the moments of “change of the polarity of the output signal”.

Strictly speaking, it is impossible to talk about a certain range of switching frequencies of the output stage arms. the frequency of these moments is not determined in advance (depends on the phonogram), but for our reasoning it is more convenient to use this terminology: let it be in the range of 2500 … 6000 s^-1 (Hz). It should be noted that this frequency domain corresponds to the maximum hearing sensitivity. We also note that the level of the output signal of the UMZCH (within the framework of the acceptable loudness of the sound of the complex) does not greatly affect the moments of polarity change, or does not affect at all.

All the data that I give in the tables and in the text are based on certain calculations using a PC computer. The total volume of multiply programmed PCM audio files (samples) was about 15 Gigabytes. To improve the accuracy of calculations at low absolute levels (when manipulating the frequency response), the audio data was converted into 32-bit form without changing the sampling frequency and / or digitization in 96kHz / 24bit format (actual 18).

So, everyone’s favorite “music signal substitute” – pink noise has an average frequency of signal polarity change of 8700 s^-1 (on the segment RSM 44100/32 with a duration of 300 seconds – 2609200 transitions through 0). Thus, pink noise of constant amplitude and not limited from below by the roll-off frequency is 1.5–3.5 times harder than any musical signal from the point of view of analyzing switching distortions in the UMZCH.

How can you reduce the number of signal transitions through 0, and therefore reduce the number of switching distortions in the UMZCH? Probably – to reduce the amplitude of the high-frequency components of the amplified signal. To assess this prerequisite, we will conduct a study of signals with a decrease in the HF components according to a certain law.

The simplest way is to correct the frequency response with a linear RC (LR) circuit of the first order with a slope of -6dB / octave. However, the application of correction over the entire operating frequency range of 20–20000 Hz is impractical because will lead to a significant “loss” of the output power of the UMZCH and / or a deterioration in the signal-to-noise ratio.

It was experimentally found that the lower limit of the beginning of the frequency response decay does not make sense to choose below 50–70 Hz. The dependence of the results (the number of transitions through 0) on the upper boundary of the “sloped” AFC, i.e. the frequency of the transition of the frequency response from oblique to horizontal is of certain interest.

I carried out the corresponding experiments on counting the number of transitions through 0 different waveforms at different values ​​of the upper frequency F at the end of the frequency response decay at the UMZCH input. The results of the “several times” decrease in signal transitions through 0 signalogram before and after the frequency response correction are presented in Table 1 and in the graph in Fig.1.

Of particular interest are the data given in column 6 of the table: “Pink noise + HPF 12x12Hz + level control“. This is the original signal of pink noise (column 2), the averaged frequency and level characteristics of which are reduced to, respectively, the averaged frequency response and the level histogram of the signal of Diana Schur’s records (column 5). The period of change in the amplitude of the pink noise (modulation) was chosen 600 sec or 0.0017 Hz.

And the most interesting is the last column of the table, where the signal (column No. 5) was studied, in which the 1.25 Hz frequency “bluntly in the forehead” was raised in level by + 76 dB (quality factor of the parametric correction circuit Q = 2000). In the frequency band 15–20000Hz, the signals No. 5, No. 6 and No. 7 are the same by the averaged amplitude and spectrum.

A few words about the subjective side of the analysis of the studied musical fragments. The recordings of Zemfira and Space seem to be bright, tough, assertive by ear. The only difference between the two is the way DDD and AAD are written, respectively. Diana Shur is a typical representative of the jazz direction and the sound of her recordings (DDD) is open, free, measured. However, during a comparative listening to audio CD with phonograms corresponding to columns of tables No. 5 and No. 7, more dynamic sounding of recording No. 7 was noted, in which pauses between passages were more clearly localized. Isn’t this a “manifestation” of the VPS (virtual constant component), “partially restored” in record # 7?

It should be noted that attempts to restore the UPL by an oscillatory circuit (meaning practical implementation) tuned to a certain infra-low frequency is …