Did some measurements recently to check out the implementations of various reconstruction filters to take a look at how they affect the analog output. Fortunately I have dacs on hand that have some variability in their filters to take a look at this.
Some subjective discussion on various filter types and especially Han's thoughts about the various filters of a Gamma 2 he tried are here.
http://www.changstar.com/index.php/topic,1535.msg40729.html#msg40729Personally I have little to say about filter sounds except that I find the Ipod4g Touch's audio performance to be more than reasonable for my uses. The products themselves are too far apart in how they are built and their analog performance to provide any useful subjective opinions on filtering.
Keeping in mind the 0404's ADC limitations with rising noisefloor with increased bandwidth, I thought the measurements I did with the DACs on hand were useful presentations of their filter performance.
Methodology used can be found in JA's article:
http://www.stereophile.com/content/arcam-fmj-d33-da-processor-follow-april-2013Test Regime:
DAC sample rate: 44.1khz ADC sample rate: 192khz
A spectrum of -3dbFS white noise and a spectrum of a full scale 19.1khz test tone was combined to form a graph. 19.1khz being very close to the upper limit of the 44.1Fs bandwidth, resulted in an aliasing product at 44.1-19.1=25khz.
White noise is played to simply observe the rolloff characteristics of the DAC's filter.
First up is the Ipod4g Touch which uses minimum phase filters in a sharp rolloff, suppression of the aliasing product is not complete but reasonable I reckon. The Ipod touch might be using some inband noise shaping to provide better measurements within the audio range. With the white noise, some rippling can be seen in the stopband, this is more of a filter design decision(high order but lower rejection) and not a limitation of the analog circuitry.
Next the Audio GD NFB2 which uses a linear phase slow roll-off filter. Slow roll-off means little rejection of aliasing products. AGD having no feedback, harmonic products of the 19.1khz tone and the 25khz aliasing product subside into the noise floor as frequency goes up.
Last is off course the ODAC with its traditional linear phase brickwall. Excellent aliasing rejection of ~-118db. Unlike the AGD harmonics go all the way past the measurement bandwidth.
Forgot to put in the wideband noise measurement for the 0404 as a baseline reference.
Disclaimer: Maybe you can hear it, maybe you can't? LOL YMMV, not Science, BS.
