Yes if the signal is hot there are a lot more samples near the maximum that “guard” the bad normalization. However with a signal with more dynamic range there are shorter impulses and when an impulse is not sampled exactly at the peak you can have problems: the vertically offset sinc function is the limiting example. 3dB is a good practical rule of thumb.
Put another way: your intersample peak can’t be higher than the gain on the normalization 
Thanks, Ted! I greatly appreciate it.
The emotional tension for me is accepting I need not slam the meters as one tends to do with tape, but if I do not record hot the concomitant need to boast volume/normalize is gently scary.
I like to believe overthinking provides mental exercise.
Overthinking? Here’s over thinking:
Sample a sinc at the red points, you get all zeros except for a 1 at x = 0.
Sample the sinc twice as often and then offset and scale the min and max to be -1 and 1, then throw away the original sample points (green) and the red points are left.
The original sample’s gain is +5.3725dB
Ted, Did you happen to have that graph handy from your development of the new DS firmware? 
Wild stuff.
Yep HQP tells you on the fly if it’s clipping and you can adjust on the fly too…
I suggested leaving it on -6dB if you’re like me and have no interest in staring at the HQP clipping indicator every track.
But sure if you know it’s a well mastered track, you can go to 0dB (or -2dB or whatever) on the fly easily.
The issue that I’m having is that with HQP on Mac Mini set at 0dB I’ve got to crank up the DSD to 90 for moderate background levels - not enough headroom left for louder sessions.
HQP is the only SW/HW combination I’ve tried in several years that has had this issue - everything else I’ve ever used lets me run the DSD set at 50-80 instead of 90-100 (and still not enough). With Roon Bridge I can set both Roon and the DSD at levels in the 70s and 80s without issues.
Since it doesn’t seem to bring any (non-signal-bent) SQ to the party, I guess I’ll be using Roon Bridge instead.
I use HQP for years and I just recently set it to Odb with good results. I always keep an eye on the limiter, but as long as I play classical music I’m fine. If I have to play Pink¡ I use the TV speakers…
I had a similar plot for a post I made about proper interpolation so I couldn’t resist doing it: here’s the code in MATLAB/Octave:
First plot:
xxx=-4:0.01:4;
plot(xxx, sinc(xxx));
stem(-4:4, sinc(-4:4));
Second plot:
xxx=-4:0.01:4;
o=0.5;
t=sinc(o);
b=sinc(3*o);
20*log10(2*(1-(t+b)/2)/(t-b)) /* The gain calculation */
plot(xxx, 2*(sinc(xxx)-(t+b)/2)/(t-b));
stem(-o*5:o:5*o, 2*(sinc(-o*5:o:5*o)-(t+b)/2)/(t-b));
stem(-4:4, 2*(sinc(-4:4)-(t+b)/2)/(t-b));
You know I was fishing right
Yep, but we should stop hijacking this thread
Can some share their latest experiences with HQPlayer?