Part 2 ... for insomniacs everywhere
Summing up some of the afore-mentioned issues with analog tape:
Limited high frequency response because of the physical size of the head gap and magnetic wavelength on tape
Limited HF response due to record EQ, and bias self-erasure
Limited HF response, particularly in playback on a different machine than recorded the tape, by relative play head misalignment, particularly azimuth.
Limited LF response, inversely related to tape speed. Higher tape speed = less LF extension. There is no recording 5hz at any speed.
Let me expand on that last one just a bit. The process of recording (magnetizing) and play (the inducement of a current in a coil of wire by a fluctuating magnetic field is an AC coupled system. That means you can't record DC, and the output of the system decreases with falling frequency. The slower the magnetic flux change, the less output you get out of a tape head. Translate this to the magnetic wavelength on a moving tape, and you'll start to see why LF response falls apart with high tape speed. The magnetic wavelengths on tape are far too long to induce much current in the head. You can equalize that response back to a point, but there is point of diminishing returns beyond which you can't recover LF response. There is a huge problem with calibration here too. LF test tones on test tapes have been historically inaccurate. One studio client of mine used one test tape to equalize the general response, another from a different manufacturer to adjust LF eg. There's also the "fringe effect", where a recorded track that is wider than the track width of the play head causes elevated (and inaccurate) LF response (called a "head bump") that is partially predictable, but counter intuitive. And that's just the play side.
So when I say there's no recording 5Hz, you can almost quadruple that frequency and be just as right. There's no recording 20Hz accurately either at 15ips, certainly no chance at 30ips. Oddly, 7.5ips can do a pretty good job at 20Hz!
Another point on tape, track format. I've worked with everything from full-track mono to 24 track 2", and everything in-between. 8 track 1/4" was a LOT of fun. Track width affects signal to noise, LF response, head alignment tolerance, and a few other things. Ideally, you'd be playing the tape on a machine that exactly matches the track format it was recorded on. In the consumer world, this can be an issue. We had 4 standards for 1/4" tape, 1/2 track mono, 1/4 track mono, 1/2 track stereo and 1/4 track stereo. None are really compatible with each other…you have to do at least something just to get usable playback. And usable isn't optimal (except playing 1/4 track mono on a 1/4 track stereo deck, that works fine).
Dynamic range? Well, it's dependent on tape formulation, and not all are created equal, track width, and speed. Higher speeds move tape hiss up higher to a less audible range. Wider tracks allow more magnetism to be presented to the play head, so lower noise. Higher output tapes have more ability to record at high levels with less distortion, but what happens to all tapes and all machines is, the higher the level, the more distortion, with IMD happening quite quickly. The maximum record level on tape is frequency dependent. So if we insist on fairly clean and undistorted peaks, the typical dynamic range of tape is around 70dB (about 12 bits). If we add Dolby A noise reduction, we can push that to 80dB (about 13-14 bits). Dolby SR stretched tapes dynamic range to right around 16 bits, but was a professional-only system (as was Dolby A). 1/4 track stereo tapes have a dynamic range of around 60dB (10 bits).
So, we've got limited HF response, limited LF response…what are we seeing on that Audacity spectrum? In the LF region, it's basically noise, but will look like modulated audio. LF noise in analog tape is caused by several things, one being the "calendaring" or precision finishing of the tape surface which relates to intimate head contact. Slight variations there will cause a bit of LF noise. More significant is the effects of DC magnetism caused by magnetized machine components or DC currents through any of the heads. Remember I said you can't record DC? You can't, but trying causes a very low frequency groveling noise, which can also be modulated by audio. That stuff has the potential for sub-sonic content, and that's what's being seen on the spectrum at the LF end. There's a whole discussion on demagnetizing tape machines to be had, just not today. Short story, people who do it all the time probably make it worse than people who never do it at all.
Spectrum over 25KHz is almost certainly distortion products caused by nonlinearities of the tape itself, the heads, and the electronics. The third harmonic of 8KHz is 24KHz, recall. Someone will argue that audio can have ultrasonic harmonics, and that's true, but we can't really record or play them, certainly not above the limits of head gap and bias erasure, without doing something non-standard. Interesting thing about non-standard, if we changed the standardized parameters we are now locked into with magnetic tape recording, we actually could improve performance in all areas, including frequency response. But we literally have to break ALL the rules to do it.
Vinyl
The entire vinyl/LP system end to end can be considered a flat, wide band medium so long as we realize that "flat" and "wide band" are highly level dependent, and ultimately have physical limitations imposed by the size and shape of the cutter stylus, resulting groove, and play stylus. At the high frequency end, there is a physical limit imposed by the contact surface area of the stylus on the groove wall. Tiny wavelengths eventually are smaller than the contact area, and simply get bounced right over (actually creating frequency down-folding imaging). The higher the frequency we try to deal with, the lower the maximum modulation that can be handled by the system. So, those ultrasonic 35KHz signals we think the system can handle must actually be very low in level, or the results of distortion and not real audio at all. An example would be the FM carrier used in CD-4 quadraphonic discs, which was centered at 30KHz, and contained modulation products up to 48KHz. A special stylus was used, and the level of that carrier was quite low. It also didn't matter too much that the recovery of the carrier was laced with distortion, because it was frequency modulated and hit a demodulator that was relatively insensitive to a distorted carrier.
So what are we seeing on that spectrum analysis? Signals that come off a phono cartridge are in fact full of ultrasonic stuff. Mistracking, where a stylus does not remain in full contact with a groove wall, produces a lot of ultrasonic as the stylus repeatedly slams into the groove wall then gets thrown away from it. Those signals will be modulated, like audio, and in the ultrasonic range, but it's not really audio itself, rather a form of distortion product. Again, high frequency distortion, odd order, places the third harmonic of everything above 8KHz well into the ultrasonic range. And here's the point: Actual harmonic content in audio is there too, but since most LPs were mastered on analog tape, those harmonics have already been reduced or eliminated before mastering to disc. What we are seeing is the result of issues in the mechanical system, not audio.
Moving to LF response of the LP. High level low frequency information has a maximum limit defined by the ability of a stylus to stay in a lateral groove that has a lot of physical excursion. Compliance is a factor, as is the mass of the tone arm, but sub-sonic bass is actually challenging to recover from disc. For one thing, the disc itself is full of recorded rumble, and even the best turntables also have quite a bit of it. Adding high levels of subsonic bass to that rumble can, and does, clip preamplifiers. The maximum bass level is also a function of pan position. You can record quite a bit of level into a lateral groove, but not as much in a vertical groove. The vertical limit is defined by the maximum depth before the cutter hits the aluminum substrate, and the minimum depth before the groove becomes too shallow to hold a stylus. Left or Right only signals are effectively a combination of vertical and lateral groove modulation, so they have those limits imposed. It's one reason loud records have bass and kick drums mixed center.
So can we get a bit of extreme LF out of a record? Sure, but how did it get there? Not from tape! There were a precious few direct-to-disc recordings made over the years where a live stereo mix was fed directly to a lathe, and those have some pretty amazing signals on them. But the "mask" of vinyl is essentially the same "mask" as analog tape, plus some new physical limits.
Vinyl has a signal to noise ratio of 50 to 70 dB depending on the original tape master, the type of vinyl used in pressing, and the condition of the record, and method of measurement. 50dB is equivalent to just under 9 bits.
So, to effectively capture everything on analog tape and vinyl, we need to exceed those medium's capabilities. 16 bits at 48KHz accomplishes that more than adequately. 24 bits (remember, it's not REAL 24 bits, it's a REAL 18 - 20 bits!) is substantial overkill, but if there's a desire for post processing, de-clicking, noise reduction, eq etc., then it's nice to have a few extra bits to play DSP games with. All high-frequency needs are easily met by 48KHz sampling rate, but if we wanted to give in to paranoia, the next rate up will approximately double that limit, so 88KHz would be far more than we need, but the next common step. And it's not as common as 96KHz, so that's more likely what would be used. 24/96 would be the choice for the paranoid, so long as you ignore the ultimate "release" format, probably down-sampled so it can be universally played. Resampling never improves audio, not in either the up or down direction. It would be best to digitize in a format that could actually be used.
How paranoid is digitizing a 9 - 13 bit equivalent signal to 24 bits? How paranoid is digitizing ultrasonic distortion products and noise? That's a deep-seated psychological question out of my area of expertise. There's no "engineering" reason to do it though.