13 - WNL-WD EVP Receiver - by Andrés Ramos

[Andres Ramos]

1. Abstract

This article is referring to my previous paper "12 - Time Domain Noise level Discrimination - an approach to enhance ITC voice modulation". In this paper I hypothesized that the bad voice quality was related to the spectrum of germanium generated noise and could be improved by using white noise.

I wanted to prove this hypothesis and I revised my electronic circuits to employ white noise from a Zener diode to compare the spectral composition and the voice quality against my results with germanium noise.

 

2. Electronic schematic

 

You see that compared to my previous circuit not that much has changed. The only difference is the noise generator. I now took advantage of the noise generated by a 3.3V Zener diode that is emitting very clean white noise. Since the noise amplitude is smaller than the one generated by a germanium diode I needed to implement another 20dB amplifier stage. I also found out that an adjustment for the output bias of the last noise amplifier stage is useful to make the signal more symmetrically. The rest of the circuit stayed unchanged.

 

3. Test results with white noise level discrimination

There were no noticeable changes in the time domain signal of the noise. However it was noticeable that the post processed voices sound more rich and not quite as croaky. The modulation was much better compared to the results of the previous setup with germanium noise. This is no wonder because the modulation enhancements rely on the signal discrimination and not the noise itself.

Time domain representation of post processed voice signal

 

The next step was to compare the spectral compositions between the previous setup with germanium noise and the new setup with white noise.

Spectrum of post processed voice signal with germanium noise

 

Spectrum of post processed signal with white noise

It is obvious that the signal based on white noise has a quite richer spectral composition than the one with germanium noise. A second reason for the higher bandwidth of the latter lies in the fact that less post processing was needed with white noise, i.e. denoising and filtering. All I did was apply Paulstretch with a stretch factor of 1.6 and 0.1s resolution, followed by 23dB and 7dB denoising.

To make the differences between the old design with germanium noise source and the new design based on white noise better comparable I have put a 10s audio sequence of each signal into one mp3-file. The first 10s are with germanium noise and the second with white noise. Both are separated by a 2s pause.

Comparison of sounds

You can find some additional audio exports here.

In a final evaluation the conclusion can be drawn that white noise really does improve the signal quality in noise level discriminated signals. This is not only mirrored by the spectral composition but also by listening to the exported audios. Thus this experiment is valued as successful.