Skip to content

A Closer Look at Oversampling & Noise Filtering: It’s Not Magic!


Oversampling is a crucial part of a modern DAC, especially the ES9038P which uses 4x oversampling in stereo or 8x in mono mode. The ES9038P has shown its ability to filter out noise that the Play allowed through. Now we know this digital noise has been brought on by having 2 paths to ground, one in the USB cable and one through the chassis ground (supplied by the power supply ground). Oversampling helps to push noise/distortion outside of the audio band, especially with modern filtering/shaping on top of oversampling. With specific shaped filtering you can filter out 50% more or so of the quantization noise than with a plain filter (Source: Art of Digital Audio).



In simple terms oversampling means to sample at a higher rate than is required by the Nyquist Rate (a little over two times the top input frequency). Quantization (signal processing) error happens when going from analog to digital (ADC). This involves taking a continuous signal and segmenting it into “boxes” or “bins”. “Such errors are an inevitable result of the classification of continuously varying analog voltages into discrete digital bins. This quantization necessarily adds this noise at the level of ½ LSB (Least Significant Bit).” Pg.694 Art of Electronics Lab Manual 3rd Edition Luckily for us, the entire process is digital, from the hard drive to the XMOS and then to the ES9038P (or ES9018K2M for the Play), where digital is then turned to analog (DAC).

Courtesy of Yamaha

Often after conversion from digital to analog (DAC), flat lines will have a roughness, also called “steppy edges”. This error can be smoothed out using a quality, low-pass filter. If you look on the back of the Play, this is what the “LP” means in the diagram. A common use of discrete op-amps, includes being used as high quality, low-pass filters with little effort.

Courtesy of Yamaha

Sampling Rate: In simple terms, sampling involves breaking the waveform/graph into smaller and smaller chunks. However, the sampling rate must be taken across the horizontal/time axis. Note: In audio, when signal frequencies begin close to zero, the highest frequency simply is the bandwidth. This tells us the sampling required for a given signal must be dependent on the highest frequency in the analog wave. If the wave is made up of more than one frequency  (such as in audio), it is the highest frequency that we are concerned with (lower frequencies are much easier to sample). So, if you buy a high res PCM/FLAC album that is rated at 96 kHz, the highest frequency you should see on an oscilloscope of the DAC outputs should be right around 96 kHz.

Prev Page

Next Page


Table of Contents:



Copyright © 2018 Hallman Labs™, All rights reserved. No material written or photographed on this site may be replicated/copied/quoted/etc. without written permission from the site author/owner of Hallman Labs.


%d bloggers like this: