What on earth bit me?????

[Jim Andrada]

I recorded a local orchestra Saturday doing the Tchaikovsky 5th Symphony.

I've recorded/videotaped the same orchestra in the same venue with the same gear half a dozen times. but this time something leaped out of the woodwork and really bit me in the ...

Here's a clip (Be careful - the interference is LOUD)

https://www.dropbox.com/s/0avh3eh6q3ypvfg/Noise%201.wav

I've never had anything like this happen before.

Setup was a pair of Schoeps mics running into a Sound Devices 702 via Canare StarQuad cable. Cable was well taped to the floor and cable covers were in place where the cable crossed a walkway.

One thing different was that this time the 702 was in the balcony instead of up front next to the mic stand. Altogether there was a bit over 175 feet of cable between mic and recorder. The 702 was plugged into an electric socket instead of being on battery power as I usually do.

I have another recording from a children's concert a couple of hours earlier in the same space as well as a backup recording on a Sony PCM D50 so between these and Izotope I think I can recover, but it is pretty bad. There are also 4 or 5 other similar hits but each is under a tenth of a second so more easily repaired I think.

The real question is what was the source of the noiseand what I need to do to avoid it in future.

Thanks

[Battle Vaughan]

Wild stab at this, so others more expert may have a better idea. Sounds like white noise, like from an open input, such as you might get if the signal dropped off from a bad connection in a connector leaving a sensitive input running open...but.... Arguing against that is that it's very loud, and that hints that it is sourced somewhere. Failing preamp, perhaps?

It doesn't sound like the 60-cycle hum you get when you have bad shielding so I'm thinking the ac power is not a factor...

[Jim Andrada]

Thanks Battle. This really has me stumped - if it were later in the Summer we'd be having a lot of lightning, but skies were clear - no lightning to be seen. And the noise burst was pretty long.

And as you said the fact that it was so loud and sustained seems to argue against a loose connector - it had to be coming from somewhere. Nobody was welding in the next room or anything like that.

[Colin McDonald]

One of these by the sound of it. :-)

Wow, that is some hit! Something out there really hates bassoons.

Thanks for the warning - even so, I looked before I listened:

EDIT: the fact that it hit both channels tells us something (just not quite sure what).

[Dave Blackhurst]

I've worked with audio for a LONG time, and that is a strange one! I might agree with Colin that you've got a Dalek infestation, but...

Since it's a hard wired run, that likely rules out signal interference, unless you were perhaps crossing over some other wiring that had issues, but then the way it's only short spikes sort of rules that out. Doesn't sound like any A/C spike or interference I've ever heard... yet the A/C being one variable does lend to that possibility. Doesn't sound like a motor or tool feeding back, not sure WHAT would create the sudden noise!

Whatever it was, it certainly was strong, indicating something probably quite close that caused electrical interference that was not audible except to your gear. It sounds generally like pretty random high freq hiss, for want of a better description...

I've heard Cell Phones cause something vaguely similar when in close proximity to other equipment, and those are of course common anymore, but I would think nearly direct contact with your signal chain would have been required, and it shouldn't be so abrupt.

[Richard Crowley]

I would bet substantial money on your storage media. That does NOT sound like acoustic or electronic noise to me. That sounds like the random "bits" you get from a bad segment on your CF card. If that is a new card, I would send it back. If that is an older card, then it has developed a bad segment and I would NOT trust it for any further use in recording live performances. The card is "supposed to" identify bad memory blocks and bypass them "transparently" to the user. But when there are too many (because the card is dying), or something happens to the internal card management, you get a segment full of random bits. That is exactly what I think I am hearing here. This is why we have backup recorders for critical situations (where the budget permits).

That is NOT any kind of cell phone interference I have ever heard. The reason it is so loud is that random ones and zeroes tend to produce Full-Scale "noise" samples. If you look closely at the "waveform" in a DAW, you will probably see a nicely-defined block of full-scale random samples, after which you are returned to your regular programming without further ado.

[Jim Andrada]

Thanks guys!

Yes, this is one of the strangest F--- ups I can recall!

I did have a back up, although not the same quality. I think the phrase that got clobbered repeats often enough that I can find another copy - in fact the Izoztope "Pattern" repair tool seems to find a phrase that fits well enough that only the soloist might notice.

I was actually feeding sound out to three cameras for sync and to an Atomos recorder as well via HDMI so I'm going through them now - if it was a media hit I would expect that the other-in camera copies would be pretty clean. In fact if so it would align with Richard's cogent analysis. It was definitely a big square block of noise.

[Greg Miller]

I admit that Mr. Crowley's explanation sounds very plausible.

Yet when I zoom in on the waveform of the noise, I see distinct repeating patterns. Some sections look like a sawtooth waveform. Other sections have a rounded nature, sort of like the top of a sine wave (only, of course, they are not nearly as pure in form as an actual sine wave). These two characteristics, sawtooth and rounded, seem to repeat throughout the entire noise burst. I don't think a section of purely random numbers would produce these visually identifiable patterns.

Also, if I look at a spectral analysis of the noise burst, it is very flat, with just a slight rise from 20 up to around 3,000 Hz. Then there is an obvious rolloff, about 6dB, as you go from 3kHz to 24kHz. This is not what I would expect from random white noise.

So I still suspect some sort of equipment malfunction. Is you equipment "RF hardened" and to what extent? How much field strength can it tolerate? Suppose someone keyed a 10 watt FM transmitter (e.g. campus police) very close to your equipment ... would it tolerate that? Or suppose there is some sort of carrier-current telemetry in use in the building for security, lighting control, temperature control, etc... would your equipment tolerate a lot of RF leakage coming in over the power line?

Back in the days when line-powered reel-to-reel was the only option, I did a lot of musical recordings on the Penn State campus. I often heard sequences of audio tones of various pitches in my headphones while recording (and, alas, on the tapes). Eventually someone revealed that the university was generating and transmitting these audio tones -- in the range of a few hundred Hz -- over the entire campus power grid, for the purpose of turning on and off various outdoor and building lights at night. I had to insert a special 60-Hz lowpass filter in the AC mains feeding all my equipment, in order to get clean recordings. It seems quite conceivable to me that people are doing similar things today, but with RF rather than audio tones.

Admittedly I am just speculating, but based on the observed waveforms and the spectral analysis, I question whether this is a case of random data from a bad card.

[Jim Andrada]

Well, another piece of the puzzle presents itself

I checked the camera audio that was being fed from the 702 - and it is clean

The setup, more exactly, was, mic --> 702,702 Line out --> Motu UltraLite mk3 -->3 cams and Atomos.

And the Atomos audio tracks are clean - no blast of noise.

So whatever is going wrong must be going wrong internal to the 702's digital stage as the analog signal out seems to be just fine

[Greg Miller]

Yes, that's an important piece of the puzzle! Glad to hear that you have clean audio downstream, that you can perhaps piece in. It might be more convincing than trying to find a similar musical passage elsewhere in the performance.

Looking at the waveforms some more, I think that perhaps some of the bits -- but not all of them -- were wrong. Either there was a malfunction in A-to-D conversion, or a malfunction of the card.

It does seem significant that both channels have the same malfunction at exactly the same time.

For what it's worth, there are 109,227 bad samples. That does not correspond to any binary number of samples (for example, 131,072 memory locations would correspond to 2^17 sequential addresses). So I am inclined to think that it's either an A-to-D failure, or perhaps an intermittent connection somewhere between the A-to-D converter and the memory card itself, or even within the memory card. But assuming the card's internal geometry is binary in nature, this number doesn't seem to relate to a binary-size block.

If I wanted to take a lot of time, I'll bet I could laboriously reconstruct parts of the waveform. But it might take an hour just to fix part of one second. It might demonstrate the nature of the data error, but wouldn't be a reasonable way to repair the file.

[Jon Fairhurst]

I hear that there is a lot of solar activity right now. It's solar flares or sun spots, one of the two. Or maybe a gap in the space-time continuum.

It's definitely not your typical 60 Hz hum or cell phone carrier signal.

[Jim Andrada]

Thanks Greg

I sent a description of the problem to Sound Devices. Hopefully they'll have some ideas

I did indeed piece in a section from another device - had to use Audition, which I haven't touched since version 2 - a bit of a learning curve to say the least. However I think it will work out OK. There are a few more hits here and there but they're really short and I think the Izotope pattern tool can mask them well enough.- we shall see!

[Richard Crowley]

It was definitely a big square block of noise.
Yet when I zoom in on the waveform of the noise, I see distinct repeating patterns. Some sections look like a sawtooth waveform. Other sections have a rounded nature, sort of like the top of a sine wave (only, of course, they are not nearly as pure in form as an actual sine wave). These two characteristics, sawtooth and rounded, seem to repeat throughout the entire noise burst. I don't think a section of purely random numbers would produce these visually identifiable patterns.

That seems within the possibilities. If the block of data had something else in it before, it might show some kind of pattern when viewed as unsynchronized audio sample data.

So whatever is going wrong must be going wrong internal to the 702's digital stage as the analog signal out seems to be just fine

That seems completely consistent with a recording media failure. I have orders of magnitude more faith in the integrity of a premium Sound Devices piece of hardware than in a commodity flash memory card. Those things don't last forever. All electronic components (including memory chips) are subject to the "bathtub curve" http://en.wikipedia.org/wiki/Bathtub_curve

For what it's worth, there are 109,227 bad samples.

If you were recording 24-bit stereo, then each "sample" is six bytes. 109,227 * 6 = 655,362 bytes which is 10 blocks of 64K (a very common block size in memory). So that is 10 blocks of 64K bytes with two bytes left over from the beginning and/or end of the failed 10 blocks. And two good bytes by themselves are incapable of recording even a single correct sample when the third byte is inside the failed block. I would throw that CF card in the nearest trash can ASAP.

The patterns Mr. Miller observed may even be un-synchronized audio waveforms from a previous recording. Or even fragments of an image if the CF card was ever used in a DSLR, etc.

[Greg Miller]

Originally Posted by Richard Crowley

I have orders of magnitude more faith in the integrity of a premium Sound Devices piece of hardware than in a commodity flash memory card.

Yes, I would have to agree with that, unless there was an extremely unusual situation in the environment like a very bad RF hit. But compared to Sound Devices hardware, flash memory cards are basically disposable.

Originally Posted by Richard Crowley

If you were recording 24-bit stereo, then each "sample" is six bytes. 109,227 * 6 = 655,362 bytes which is 10 blocks of 64K (a very common block size in memory). So that is 10 blocks of 64K bytes with two bytes left over from the beginning and/or end of the failed 10 blocks.

That's a good catch. I should have gone one step further with my math. That is certainly very strong support for your conclusion that it's a bad memory card.

Originally Posted by Richard Crowley

The patterns Mr. Miller observed may even be un-synchronized audio waveforms from a previous recording. Or even fragments of an image if the CF card was ever used in a DSLR, etc.

They look very much like a damaged audio waveform. Specifically, I would guess that the most significant bits are stuck "on" bringing everything up to nearly full scale. When I look at the areas that look like a sawtooth wave, they tend to have eight sequential samples going in one direction (e.g. descending), then the next sample jumps in the opposite direction (e.g. up) with another group of eight samples going in the original direction (e.g. descending). I can easily visualize putting these together into a continuous audio signal, as in the attached image.

[Richard Crowley]

Originally Posted by Greg Miller

They look very much like a damaged audio waveform. Specifically, I would guess that the most significant bits are stuck "on" bringing everything up to nearly full scale. When I look at the areas that look like a sawtooth wave, they tend to have eight sequential samples going in one direction (e.g. descending), then the next sample jumps in the opposite direction (e.g. up) with another group of eight samples going in the original direction (e.g. descending). I can easily visualize putting these together into a continuous audio signal, as in the attached image.

Indeed, it may actually be Jim's recorded waveform, but simply scrambled because of an error in the bitstream. I'm not sure what would cause it to "recover" after 10 blocks, but I need to get some more sleep.