Bought a matched pair of NOS RCA 6SL7s from eBay for my 845 SET amp. Sound was great except for what seemed like smearing, massive smearing, at about the 1000Hz range but not on every track. Ordered new tubes a month or so later because it was unlistenable and the sound had become compressed (I may be sending them back when they arrive). Yada yada yada it turns out it was my settings.
Upsampling settings to be exact. Upsampling to DSD256 with rBrain, but had—for reasons that were associated with an issue a good 9 months ago—Nyquist at 99.3 and Stop Band Attenuation at 166, as well as using Modulator C.
After digging into setting for unrelated reason last night, I set it back to 99.5 and 218 (defaults, I think), as well as B8 instead of C. Smearing was gone, like, not even a thing anymore anywhere, and the sound is better than I’ve ever heard.
I’m assuming it was the change fro 166 to 218. Any thoughts on this?
Probably need some more info about your playback settings/system and computer platform/OS and available System RAM and playback pre-load memory allocation, plug-ins, etc…
The stop-band attenuation is probably not the culprit…
Are you attenuating the signal before upsampling? You could be experiencing digital clipping distortion…
Although I modulate to DSD128, my configuration of r8Brain has the Nyquist at 99.5% and Stop-band Attenuation at 166dB, with A-type 8th order filter… and I employ a HRTF plug-in before upsampling to DSD128… I attenuate the signal from within the AU plug-in by -4dB and leave the gain reduction setting in r8Brain at ‘none’… I have no issues with distortion in my DAC where I also employ a FIR filter…
MacBook Pro (15-inch, 2016)
Processor 2.7 GHz Quad-Core Intel Core i7
Memory 16 GB 2133 MHz LPDDR3
4GB allocated for pre-load buffer
Upsamping will increase the dynamic range of the source file, and this must be accounted for.
Mac mini M1 with 16GB memory with 12GB allocated to Audirvana. If 166 to 218 isn’t the issue, then the only other things I did was go from 99.3 to 99.5, flip from C to B8, and go from none to -3. The difference was dramatic.
I suggest lowering your playback pre-load memory allocation to 4-6GB… There is no real good reason to allocate more than this…
From the online Help Information in the Upsampling window:
For the R8Brain algorithm, you have access to the following settings:
○ Bandwidth: this indicates the limit of the low-pass filter as a % of the Nyquist frequency (half the sampling frequency). The slope of the filter is low at 74% and very steep at 99.5%. Make sure that this does not induce too much brightness (i.e. highs are too aggressive).
○ Cutoff Band Attenuation: this is the setting of the slope of the low pass filter expressed in dB per octave. With the maximum slope (218 dB), there may be too much brightness (i.e. the highs are too aggressive)
○ Phase: All low pass filters have some level of overshoot. The steeper the slope of the filter, the greater the overshoots. There are two types of overshoot: pre-oscillations and post- oscillations. The former are audible as a “pre-echo” arriving before the signal itself, and are the least natural to hear. You can choose between a filter setting with linear phase, but with an equal level of pre- and post-oscillation or a minimum phase filter, with no pre-oscillation, but non-linear phase distortion.
Tubes are famous for high levels of even order harmonics (frequencies that weren’t in the original, that are 2x, 4x, 6x…, etc., the original frequencies). Even order harmonics are often pleasing and can create a full, “lush” sound. (If you have a piano, play any note by itself, then the note one octave above at the same time, which is 2x the frequency. The effect from tubes is much less, but the more full sound of the added harmonic on the piano gives you a very general idea.)
So what’s wrong with a little lushness? Well, too much of a good thing can interact with the original signal and cause something called intermodulation distortion. When you increased the stop band filtering, it could have reduced higher frequency harmonics and reduced any possible intermodulation distortion.
There is also a phenomenon called “intersample overs.” The original digital samples will all be below 0dB, maximum volume, but the recreated analog signal might go above 0dB at times between samples, and this creates distortion. Reducing the level of the reproduced signal 3-6dB as you did would eliminate this type of distortion.
So it looks like either or both of the changes you made could have cleaned up the sound to some degree and eliminated what you heard as “smearing.”
It does appear to be a combo thing, yes. I hopefully will have time in the next few days to go back to my original settings and change them one by one to see which it actually was, but my gut tells me it was both the -3dB and the 166 back up to 218.
This statement is contrary to what is described in the upsampling help information as follows:
○ Cutoff Band Attenuation: this is the setting of the slope of the low pass filter expressed in dB per octave. With the maximum slope (218 dB), there may be too much brightness (i.e. the highs are too aggressive)
