That is the more common use for digital IIRs in audio applications, paired with convolution to serve as equalizers and/or room correction. The technology can also be used for the regular anti-alias low pass filter, although not seen as often as FIRs, mainly due to the inherent phase linearity issues and reliance on feedback.
1 Like
The HQPlayer software is the NAA endpoint for the roon core,
basically playing the role of a network transport, from there you can connect to the DAC via USB or via the network, but the latter would require the use of another NAA network transport, one additional hop not really needed. I do that sometimes to stream to an iFi Zen Stream on a secondary system, although most of the time I use RAAT for that purpose.
I know that but it’s also the computer doing all the processing and heavy lifting and a potential source for additional noise, where using a separate NAA would remove that possibility and may improve it as it’s just a simple endpoint doing less and you put the pc somewhere else out of sight. Only a thought but that I believe is the idea behind the NAA or do I have that wrong?
1 Like
It’s not so much lack of computational resources these days with FIR, it is the switching noise created by the large kernel filter matrix convolution (taps) interferes and detracts with the resolution gained… this has improved with very low current devices and very low current FPGAs but is still the limiting bottleneck as well as latency issues and sample rate dependent limitations with FIR…
IIR is theoretically significantly superior to FIR in terms of resolution and more efficient operation, but it is bound by phase issues.
Regarding Naim, as you may be aware they use Analog Devices DSP, and it’s a programming option whether to use IIR or FIR, and in listening tests I understand they have opted to use IIR as sounding preferable for reconstruction, and the theoretical phase issues have not proved to be significant.
Now if Naim moved away from SHARC they might choose a different convolution method.
I wouldn’t say FIR is generally preferred for SQ… it has come of vogue of late probably spurred on by Chords adoption and the marketing of more taps must mean it’s better, but that is different … but that doesn’t mean it’s superior… and indeed I remember a 1-1 conversation at a certain county show ground talking to Mr Watts about the limitations of FIR … not least reduced performance at higher sample rates…
Where FIR is beneficial is that it is easier to create a composite hybrid kernel, rather than having a chain of IIR filters… this is ideal for complex eq response filters… but that is of course a different use case compared to DAC reconstruction which is typically a simple sinc filter.
I think the main idea behind NAA was to create a framework for third party support. Still, most network transports run linux distros at the user level, even if heavily stripped down they would still be doing must of the things a PC does, so I would not categorically conclude that the bigger PC is necessarily that much noisier than the average appliance.
The way I see it the key is to isolate the DAC from all that noise. A quiet power supply is a good place to start, then I would be more concerned with the noisy Amanero USB interface, which is in turn receiving a noisy voltage rail that even Chord seeks to avoid with the M-Scaler dual-coax setup.
In my setup common mode noise is mitigated first by the iFi decrapifier, then by the SRC•DX itself, and airborne EMI by the bandpass filters on the Qutest. I know there are better ways of doing this but at a cost I’m not willing to entertain.
Signalyst seem to think it’s important to take the heavy processing away from the output side as they plug it on their website so I would think it’s important for maintaining higher sq but if your happy who am I to say otherwise. As for NAA most I know run them on lightweight distros on a sbc or dedicated audio device with Roon and HQplayer server on something else.
1 Like
I’m aware of the general guidelines regarding NAA, those are meant to discourage the average end user from pluging in to their USB DACs from their all purpose mac book, with their noisy SMPS, fan, monitor, wifi, all while they type emails and browse the web.
The configuration I have is quite common and have been discussed online with Jussi at length. In fact, if you have a Chord DAC and want to benefit from avoiding the noisy Amanero USB interface with the SRC•DX, Windows it is about the only reliable way of doing it, as linux kernel drivers are not well supported.
A headless, fanless, wired, linear powered dedicated PC (properly stripped down), becomes a de facto lightweight client. The so called heavy processing is mere 3% processor utilization with an i7 Core. Context is everything.
1 Like
Where did you find a stripped down version of Windows 11?
I did it myself by uninstalling unused software and disabling unneeded services. You may want to look into Audiophile Optimizer if Windows 10 is okay. I didn’t go that route because I needed the network services for remote administration and file system management.
There was/is an AME version of Windows 10, which I tried for a while. Thank you for the informatiom
1 Like
There is no such thing as a convolution filter - convolution is the method of multiplying a matrix of an FIR filter kernel (ie it has finite size) over a moving window matrix of the digital series signal. Convolution is a mathematical / DSP term - it means effectively multiply together. Yes Roon uses FIR filter, typically for its room matching filter as well multipole parametric eq filter which is an IIR filter
Absolutely - and that has been standard DSP orthodoxy going back to the 80s
Thanks for the technical explanation. 
Roon uses the term “Convolution Filter”, it probably makes things easier to understand for the general user rather than the Roon team getting things wrong.
Similar to the vehicle part often referred to as “brake disc”, there’s no such thing, it’s a brake rotor (usually with an integral ‘bell’) that is part of a disc braking system. “brake disc” is now universally used outside of engineering establishments.
Of course it is a disc, unless you are ‘ merican.
2 Likes
indeed - in simple speak I guess we have a filters that use convolution with a target filter response and a filters that use procedures - but I think finite and infinite response filters to me is an easier concept to understand - but then perhaps that is just me.
A good analogy for me is a Vacuum or Vacuum cleaner - so I guess with these sort of layman short hands - context is key. The cleaner is the function - but it is achieved by using a vacuum.
1 Like
And yet Chord runs the “heavy processing” on the M_Scaler directly attached to the DAC.
Indeed the mscaler deliberately runs the heavy processing away from and decoupled from the DAC for this reason. Chord also suggest physical separation between mscaler and other sensitive components, as the case of mscaler is not a complete Faraday cage.
Yes there is a degree of coupling via the BNCs but it is has designed so I understand with physical decoupling in mind and the all important ground planes in Chord’s approach is entirely decoupled, in fact Mr Watts told me once they could see or measure no modulation on the DAC (DAVE) ground plane
Now Naim is in a little bit of a corner … as it has promoted it’s all in boxes for its current streamers, but in the they go as far as they probably can with the one box limitations to internally decouple. They use LVDS, multi layer PCBs, digital module screening, seperate Powerline regulation and physical layout… and although the current streamers are massively improved over their first gen products, reading these forum pages you can see it’s not completely successful… and physics and electronic engineering principles say there is only so far you can go with this approach.
On the purely analogue signal side, Naim do decouple the preamps from the poweramps at the higher end… though I would argue there is less need to do here compared to digital/analogue combined components.
Tried it but then dumped it… ISTR it was something to do with the Tags you could search on. It’s several years ago but I think it resisted searching on AlbumArtist (which is my preferred base tag). Also, if you only use local streaming from a NAS (or similar) it doesn’t seem to offer anything extra - IMO 
The M-Scaler is not away, it is directly connected not by one but two electrical connections, decoupled only by galvanic isolation which is not a 100% solution.
He has publicly acknowledged that ground loops through the mains can inject RF noise into the DAC.
Point here is that the M-Scaler directly connected to the DAC via dual-coax is not inherently better than a PC connected to the DAC via the same topology, at least as far as “heavy processing” goes.
The last day for Roon before the price rise is upon us.
I have been second guessing myself and won’t buy the lifetime Roon after all (the money has been spent elsewhere).
The sound is generally more detailed and improved through LMS Squeeze/Squeezelite in my system … with some exceptions where recordings are already very bright (some Jack White/Whitestripe songs come to mind).
Roon wins on interface, better information, music suggestions, etc. To help offset some of this when using other software, I often open the AllMusic website for reference while playing (which I believe is where Roon gets some of its features, along with Spotify, Apple and others). There are probably other options, but this site is worth a look.