advertisement


"Linear" DAC

H

hockman

pfm Member
I recently came across a CD player whose DAC was described as 16 bit "linear". I know it uses a pair of Burr Brown PCM61 chips.

Can someone explain in layman terms what "linear" means in this context? Does it also imply that the DAC does not use oversampling?

Thanks.
 
Just to confuse things further one of the main reasons to use a delta sigma dac is to improve on the linearity issues caused by the finite precision of resistors in R-2R dacs. So maybe it meant a delta sigma dac.
 
'Linear could have referred to two things:

LInear pulse code modulation / 'Linear PCM' (- which originally referred to not using compressed source information) and appeared on a lot of old dacs and datasheets
or
Linear, to identify early dacs for audio (as opposed to other instrumentation uses, such as Multiplying Dacs)
 
The '16 bits linear' description is in the specs sheet for 'quantization', probably doesn't mean anything more than just that. The player is the broadcast Denon DN 961FA, a cute little thing if a bit idiosyncratic.
 
hockman said:
I recently came across a CD player whose DAC was described as 16 bit "linear". I know it uses a pair of Burr Brown PCM61 chips.

Can someone explain in layman terms what "linear" means in this context? Does it also imply that the DAC does not use oversampling?
Click to expand...
I see that the player is a Denon DN-961FA (with two Burr-Brown PCM61P-L DAC chips). "D/A Conversion 16-bit linear" on the player's data sheet is not precise enough to interpret for sure, but as the player is targeted at the broadcast market I would interpret it as customers there most probably would. But those words alone are a bit limited in value.

There are two issues. What does "linear" mean; and does it say anything about the DAC chip.

Broadly, "linearity" is often used to refer to how well the output of a system tracks its input. A system is called "linear" when, for all relevant input levels, the output level is a fixed factor times the input plus a fixed (ideally zero) offset. Plot that on a graph of output versus input and you will have a straight line. Hence "linear".

Good linearity is essential for audio but it is never perfect. It is common to plot the ratio of output to input (i.e. the fixed factor above) on a decibel scale against the input on a decibel scale. For perfect linearity that will be constant and so a horizontal line, but practically there will be some deviations. For a CD player or DAC the deviations are usually worst when the input is small.

I would tentatively interpret the data sheet as I think you have above, as just proclaiming that the deviations from linearity are too small to matter for 16-bit audio data. Which brings is to the DAC chip.

The PCM61P-L 18-bit DAC chip is a resistor ladder DAC using laser-trimmed NiCr resistors that can be operated in oversampling mode but does not have to be. A quick glance at the player's service manual suggests it is not oversampled but I could be wrong. From the data sheet, the "-L" versions are not selected for best unadjusted linearity and I would not quite say they are "16-bit linear". However, there is a trim facility and the player's service manual gives the procedure to use it. So probably the player is at least "16-bit linear" after adjustment (if I have interpreted the words correctly).
 
Quick google search and Apparently another Denon manual seems to indicate that it detects Linear-PCM, whatever that is. May have to do with the source?

Also later models were marketed as super linear… there’s even a double super linear
 
From a Stereophile review:

20-bit D/A conversion
Technically, the Denon has more in common with the Kinergetics, Thetas, and Proceeds of the world than the typical Japanese player. For starters, the 2560 uses four Analog Devices AD-1862 20-bit DACs, two for each channel in push-pull configuration. These DACs are flanked by their attendant MSB trimpots, which Denon hand-trims for highest linearity at the factory.

Denon operates these high-quality chips in an interesting configuration they call "Lambda D/A Conversion"; to minimize the zero-crossing distortion caused by MSB nonlinearity, the data stream is taken from the digital filter and duplicated so there are two data streams. After adding a constant digital "bias" to each data stream, positive-going for one and negative-going for the other, the data streams feed the 20-bit Analog Devices DACs, whose analog outputs are then summed; as the bias signals are opposite in value, they ultimately cancel, but the resultant signal is biased away from the zero-crossing line, eliminating that source of distortion. The tradeoff (there's always a tradeoff) is that for high-level signals requiring the full dynamic range of the DACs, the Lambda process is momentarily disabled, the high-level signals then theoretically masking the residual distortion.
 
All wrong.

In the early days of CD "16 bit linear" was colloquial for "non-oversampling".

Thus typically Japan-inc CDPs with two or one (time-shared) 16 bit DAC chips operated at 44.1kHz and with steep analogue
reconstruction filtering, and this in contrast to Philips-style players.

From a technical POV the use of the term 'linear' here is nonsense, as they were all supposed to behave linearly.
 
Thanks, John Phillips and Werner for the clear explanation and effort in checking on the chip and service manual. I can now conclude that the DAC is indeed non-oversampling and that 'linear' means nothing more than that.

I know 'non-oversampling' DACs are currently fashionable among certain quarters. Are there particular sonic characteristics typically associated with such devices? BTW the Denon player sounds very decent, smooth with good musical flow although it is not particularly detailed or resolving.
 
martin clark said:
'Linear could have referred to two things:

LInear pulse code modulation / 'Linear PCM' (- which originally referred to not using compressed source information) and appeared on a lot of old dacs and datasheets
or
Linear, to identify early dacs for audio (as opposed to other instrumentation uses, such as Multiplying Dacs)
Click to expand...

It could also mean what Pioneer called "Legato Link". This in essence "draws a straight line" between adjacent sample instants rather than the smooth shape defined by Nyquist. As usual, claimed to sound better. I do have two of their CD recorders that do this, and they sound fine. The nominal advantage is that the system avoids 'staircase' distortion and dodges having the render intersample 'overs'. So a 'kinder' sort of error, you could say.
 
Setting aside the appellation '16-bit linear' likely came from marketing and is meaningless, if we were to try to give it meaning my suggestion is that output voltage vs. input bits transfer function be a straight line for the 16-bit dynamic range of 96dB (or 90dB to allow for LSB dithering).
Holo-May-Linearity.png

Here is an example, from Stereophile, where that transfer function (blue line) is straight for 120dB. The red line shows the slope of the blue line and there is no deviation until below -120dBc making it 20-bit linear.
 
You must log in or register to reply here.


advertisement


Back
Top