DSP Concepts has been a technology partner for many clients in the past 12 years. Our single focus is audio DSP to help our customers deliver the best audio products. The common thread for our customers is their foremost concern for high quality audio. In this blog I will talk specifically about a subtle problem that one of our customers encountered with using a free Linux open source Sample Rate Converter.
One of our customers is at the forefront of automotive technology. The founder is a visionary and risk taker that demanded perfection in every detail. Even though he was a serial entrepreneur who is simultaneously overseeing a cutting edge space transportation company and the innovative automotive company, he personally would sit in the automobile and make sure that every nuance of audio is perfected. Such is his demand on all his engineering teams.
At the outset, a decision was made to output audio at 48 KHz from the DSP. 48 is higher than 44.1 and so 48 kHz must be better, right? However most content today that is played from a thumb drive originated on a CD and is sampled at 44.1 kHz. So, it’s necessary to convert most content from 44.1 to 48 kHz. It is innocent enough to think that Sample Rate Conversion is a straight forward mathematical function. The Linux OS used on the head unit even had a free open source Sample Rate Converter built-in.
However, the audio simply did not sound right judging from the necessity to measure up to the standards of all involved. One can say that it sounded “OK” to most undiscerning ears. But to the discerning ear, this Sample Rate Converter lacked clarity and imaging. For some reason, songs played from a thumb drive (through the sample rate converter) didn’t sound as good as songs fed through an analog input (and sampled directly at 48 kHz). Though free and convenient, this Sample Rate Converter failed to achieve high quality audio requirements.
It takes special expertise to know that one cannot simply apply linear interpolation when dealing with audio applications. It takes audio experience to know that simply taking the closest two samples and interpolating them causes high frequency aliasing to creep into the audible range of the spectrum. A closer inspection in fact revealed that some antialiasing filters in the open source Sample Rate Converter was not properly implemented. High frequencies were not sufficiently attenuated and aliased back into the audible frequency range.
Unless the proper Sample Rate Converter was implemented, it would have been better to simply output at 44.1 KHz to match the audio content sample rate. So a higher sample rate is not always the best solution.