Goran Tomas

An overview of my work of transforming the DSPXmini-FM processor to the DSPXmini-FM SE (Second Edition).

DSPXmini-FM SE processor

Anybody who has some experience in processing, knows it’s VERY subjective. People hear differently, are sensitive to different things, prefer different “flavors” and textures of the sound, etc. Describing processing and the sound becomes quite difficult as well. Nevertheless, using some short audio clips to help me, I’m going to attempt to illustrate to you the sound and the thought process behind re-designing the original DSPXmini processor. As they are only short clips, an untrained ear may not perceive a significant difference. However, had you had an opportunity to listen to these processors and effects for longer time, you would have developed sensitivity and the difference would probably be much more noticeable. Keep that in mind while listening.

As with all re-designs, things usually start with the discontent of some aspects of the existing product and an idea that things could be done better… With processing, this occurs after you’ve spent a lot of hours carefully listening to the processor. After a while, you get to know its character (almost like you would a person). You can anticipate how it will behave in certain situations, what reaction you will you get for certain input. You get to know its good sides and its flaws. And if these flaws are obvious, displayed too frequently and cannot be tamed down sufficiently with available parameters adjustments, this is when you become unhappy.

My unhappiness with the original DSPXmini was with several things. Let me try to illustrate them to you:

Note: It’s best to use headphones to listen to these clips. The processor was set to 50µs pre-emphasis and the ‘AC’ preset.

1. The overall frequency balance that the processor always tried to created was too bright. Inherently in the design of the processor, there was a strong tendency to shift emphasis to high frequencies. To the point of the sound becoming too silky and strident:

   Note: After you press play, it may take some time for video to load.

   
   

2. There was an irritating “hole-punching” effect in the high-frequencies, where the processor would distort and duck the high-end on HF transients such as the words starting with an ‘s’ and other sibilants:
   

3. The overall sound was too busy, too mashed in. More often than not, you could hear the layer of processing rather than being able to listen to the music and enjoy it. Additionally, the processor was breathing too much, slowly changing the sound and the frequency balance unnaturally during the song:
   

4. The low-end texture was rather tight and constrained, rather than being open, big and adding power to the sound.

By listening to the processor, observing the meters and noticing what effect the adjustments of the parameters had on the sound, it was obvious to me that the major source of the problems of ducking, changing frequency balance and busy and constrained sound was with the 4-band limiter. After looking at the algorithm structure and trying to modify it to correct these problems and being unsuccessful in my attempts, I realized that I need to take a completely different approach. This was when I embarked on the development of the limiter algorithm from the scratch and where I spent a lot of time trying out different ideas. Most of these ideas were from my home-brew processing designs in analog. I also wasn’t going to satisfy on the first idea I had, but on the one that sounded the best and that took time. But in the end, the new limiter algorithm was born. With smart program-dependent attack, release and gating circuits and a few tricks to handle the pre-emphasis more naturally. The new limiter sounded much better and more natural to all who listened and it was decided to try it in all the other BW Broadcast processors. But unfortunately, other priorities with the development of AM processors were pushed forward and after I decided to leave the company and go back to Croatia, there was no time left to do that…

Sneak peak on the excerpt from my ToDo list during development

However, the new limiter did not solve all the problems. Mainly, there was still a lack of bass punch. And in my book, a solid, powerful low-end is the foundation of a good sound on-air! I felt my limiter could handle much more than it was being taxed with, so I ended up making tweaks to the AGC in order to change the energy balance handled by the AGC vs the limiter. This produced some results, but still unsatisfactory so another major re-design was in order – the crossover. I already had a good design of the crossover that came out of my analog research, that I knew had exactly the performance I needed here. Without going into further details, the new crossover really helped the limiter bring out some of the bass thump and low-end up in the overall sound. Finally, a distortion-controlled clipper received some tweaks in order to reduce the aliasing distortion and the transformation was finished with several more weeks of tweaking and fine-tuning of all the processing stages so that they would work synergistically and smoothly with each other.

And here are the results (a sort of before and after comparison):

Note: It’s best to use headphones to listen to these clips. Both processor were recorded in the same environment, with the same input and output settings, 50µs pre-emphasis and running the ‘AC’ preset.

1. Frequency balance:
   
   
2. Stability of the sound:
   

3. Low end texture and bass thump:
   

For better comparison, here are two longer clips with mixed music of both the original DSPXmini-FM and the DSPXmini-FM SE:

Note: You can switch between the clips by pressing the Play buttons (only one clip will play at a time). You can also position the playback bar where you want it to start playing so you can easily compare processors at different parts of this music mix. These players will play 128 kbps MP3 encoded audio, but you can download the original WAV files by clicking on the Download (down arrow) buttons.

Source audio

Original DSPXmini-FM (AC preset)

DSPXmini-FM SE (AC preset)

The mini SE is not without flaws. One thing that I wanted to do is tackle the AGC which could be better and more effective, the overall cleanliness could be better and the processor will not exactly “burn the dial” with loudness and punch… Personally, I always want to improve more and I’m rarely completely satisfied with something. But then you have to consider the fact that this processor costs only EUR 1200/USD 1500 (at the moment of this writing) and there’s only so much development you can afford, with a profit from the sales at that price point.

However, I believe the mini SE really matured from the original processor and improved on most of its previous weaknesses. Aside from comparing the processor with its competitors, the way I judge whether the processor is good, is to see if I can listen and forget about the processor and just enjoy the music… Whatever the processor does it has to be natural, in harmony with the audio and it must not detract my attention by getting in the way and doing something unnatural. And I believe I was able to achieve that important goal with the DSPXmini-FM SE.

The DSPXmini-FM SE was presented at the IBC 2008.
Visitors could compare it to the original DSPXmini-FM and other processors.

This is the paper I presented in Graz, at the 3rd Congress of the Alps Adria Acoustics Association in 2007. It answers the question of what happens to the audio quality as you push the audio processing to the maximum, while you are broadcasting digitally (meaning there’s some kind of perceptual coding after the audio processor).

PERCEPTIBILITY OF AGGRESSIVE DYNAMICS PROCESSING IN DIGITAL AUDIO BROADCASTING

Abstract: Radio broadcasters have been pushing the loudness of FM processing for years in effort to attract more listeners, despite all accompanying negative effects on audio quality. It is likely to expect that such practice will continue with digital broadcasting as well. Our goal was to determine how listeners perceive varying degrees of aggressiveness of dynamics processing for digital broadcasting, and in particular how aggressive dynamics processing influences quality of perceptually coded audio. A listening test was conducted with eight samples of varied aggressiveness of dynamics processing, before and after encoding with HE-AAC v1 at 48 kbps stereo.

Key words: dynamics processing, digital audio broadcasting, audio quality, listening test, perceptual coding

My last article for Radio Guide in Jan 2007. explains why you need audio processor specifically designed for digital audio broadcasting, and what are the different requirements between analog and digital transmissions in regards to audio processing.

Abstract: With IBOC (aka HD Radio) gaining momentum, many stations are deciding to include a second channel (HD2) and even third (HD3) as an alternative and/or additional programming to their main channel. Many broadcasters are also looking at streaming to the web as royalty fees are now clearly regulated and more people are using on-line multimedia services than ever before.
When it comes to processing for these new services – especially if they have different programming than the main channel – it is tempting to use your old FM processor that has been sitting as a backup. But that would be a mistake. In this article we will try to explain why and what are the requirements and the differences in processing for analog FM and digital channels.

Key words: digital broadcasting, HD Radio, IBOC, digital audio processing, dynamics processing, clipping, look-ahead limiting, pre-emphasis, codecs, perceptual coding

Article on Orban Optimod PC 1100, a professional audio processor for digital broadcasting and streaming. Featured in Radio Guide in April 2006.

Abstract: To have a great sounding web stream you not only need a good codec, but you also need quality audio dynamics processing. Users have been accustomed by terrestrial radio broadcasting to set the volume at a certain level and not have to adjust it for every song. They expect the same from radio delivered through the Internet.

Key words: streaming, dynamics processing, perceptual coding, Orban, Orban Optimod, PC-1100