Project "Groove Scribe"- A DYI stereo 45/45 head.

[EpicenterBryan]

The relationship between velocity, excursion, and frequency should be:

E = V/(2 x pi x f) or V = E x 2 x pi x f

Where V is the velocity and E is the excursion. Note that the 5 cm/sec reference is the rms value not the peak (rms is what your voltmeter reads as well). The peak 1.41 x RMS (Square root of 2). So for 5 cm/sec rms, you really have 7.07 cm/sec peak velocity. If you put that into the first formula, you get E = .001125 cm peak excursion. Peak to peak would be twice this value or +/-.00225 cm = +/-.0225mm..

Aha! RMS velocity and the old 1.414 got me again!
Thanks for the equation Mark!

I’m going to need to do new calculations. The plan will be to update the diagram with correct values and equations shown so that future readers don’t go down the wrong path. Garbage in, garbage out…

As far as that new test jig goes:

As it is now, there are opposing drivers. They are 1 foot apart. The test fixture is non magnetic (aluminum). The connection between the drivers (right now) is also non magnetic Carbon Fiber. There is a support piece in the center that is also non-magnetic.

The idea here is that I can drive one side, and measure on either side knowing there is no significant magnetic coupling between the two ends. Also, driving one side will result in a mechanical load being applied on the other side. If the load offered by the spring only on the opposite side is not sufficient, the driver on the opposite side can be fed with an out of phase signal with adjustable intensity to increase the load on the opposite driver.

The push rod between the two drivers can easily be replaced by one that has a section of magnetic (like spring steel) coupled to a section of nonmagnetic material (like carbon fiber) to simulate only ½ of the 45/45 drive section in the head while still proving the ability to apply a mechanical load.

The test fixture allows mounting a measuring device with different orientations and distances to the driver or push rod using mounting adapters as needed. The fixture also allows shielding to be inserted, formed, oriented, and the effects measured.

Even though the last round of calculations were brutally flawed, it did clue me into another issue with this head design. In the diagram shown earlier, the driver connection noted as “D” implies the two 45/45 drivers are connecting at a single point (as they were in my other head design). On this head, the two drivers connect slightly offset at the torque tube. Not by much, but enough to explain why the sine wave cut is not symmetrical. Here is an earlier shot of the construction to refresh your memory... I'll measure it but I think it was 1.5mm so the two drive push rods would clear each other at the "X marks the spot".

IMG_3114.JPG

And as Mark pointed out as something to watch out for is significant levels of odd harmonics showing up when doing a 1khz mono test. I remember seeing that in a 1K test that I may not have posted in hindsight... I also saw a non-symmetric looking waveform and didn’t make the connection other than that possibly my amp must have an imbalance (which it also does). All is not lost… I just need to drive one side hotter to make sure the cutting movement is balanced.

So here again is a photo of that 1khz groove from a post from Dec 19th of last year and none of us spotted the non-symmetric nature of that waveform! Shame on me for not making the connection to Mark's comments... But shame on you guys for not noticing either! Look at the groove. The first portion of the waveform is shorter than the second part. Minimum to maximum is shorter than maximum to minimum...

IMG_3161.JPG

So I’m going to open this up to discussion until I have time to play with this….

What do you guys think?

[opcode66]

I know exactly why this is happening. I can actually explain it in detail.

[EpicenterBryan]

I know exactly why this is happening. I can actually explain it in detail.

Todd, please share when you have time. I for one do not want to distract you from BladeRunner! Groove Scribe can wait, but I CAN'T WAIT for another BladeRunner Update!!!!

[EpicenterBryan]

I'm getting back to Groove Scribe this weekend! Also doing more Caruso Massimo debug and checkout.

I'll also be shooting a video for a 45/45 stereo head demo gizmo that should help people like me (who are visual) understand the movements required to make things work, and the parts that make the movements possible. This 45/45 visual movement simulator uses RC servos and an arduino processor to run it. Oh, I had to throw in an analog pot with an old school Ampex volume knob to control what program runs during the demo. How cool is that? And since I needed to mount things, what better to use than a 14" disc of failed test cuts from Groove Scribe!

IMG_3495.JPG

IMG_3496.JPG

Stay tuned this weekend for the video (I'll post it on a new thread). When complete, the mechanism will be a fun and functional "wall hanger" for my man cave. Very cool if I have to say so myself.

Bryan

[opcode66]

I know exactly why this is happening. I can actually explain it in detail.

Todd, please share when you have time. I for one do not want to distract you from BladeRunner! Groove Scribe can wait, but I CAN'T WAIT for another BladeRunner Update!!!!

Three things. First, if your sign wave is being cut unevenly, and the amplitude is the same (but the timing of each half period is different) it can only mean one thing. That is what I see in your picuture. I already identified, something is off center (uneven) between the halves of your 45/45 system.

Second, your servo example has a huge flaw. You have moving (rotating) joints at the point where the linkage meets the torque tube. That will make your demo totally inaccurate.

Third, you can read all about this and more when you buy my eBook on cutterhead design, engineering and exposing all the secrets. Documenting every step of my path. All the pitfalls. All the discoveries. Things I haven't posted or even hinted at.

Since I'm moving, you know, the bladerunner updates will be sparse. I think I'll have one final test cut video. But, then the move. The dynamic head is done. I've sorted out the last issues with it. I am nearly done with feedback as well. I just need to source new magnets of the correct dimensions and I'll be getting the proper ratio of feedback from static magnetic field vs feedback from alternating electromagnetic field. I currently get a 4:1 ratio. With new magnets it should be more like 12:1!!!

http://lathetrolls.com/viewtopic.php?f=15&t=5610

I'll be submitting heads to Chris Muth and Kevin Gray or inspection and review. If either wants a dynamic head, then they will have one in June.

Thanks for your continued interest Bryan.

[EpicenterBryan]

already identified, something is off center (uneven) between the halves of your 45/45 system.

Yep. I think one root issue is the 1.5mm offset between where the two links go through the torque tube. That means one driver pushes the torque tube slightly further than the other. I'll need to do that differently next time.

Second, your servo example has a huge flaw.

Yep, it sure does. I mentioned it in the video. I thought about building linear stages like I did for my Plasma Cutter Project below. This one has 3 belt drive, and 2 leadscrew stages. Just not enough hours in the day, and too many projects in the works man!

IMG_3497.JPG

when you buy my eBook on cutterhead design

Great idea to sell that. Put me down for a copy.

Bryan

[EpicenterBryan]

I made some progress this weekend. I sure am glad I made that driver test bed. That proved to be very handy. I spent a day messing with it and the phone cartridge / preamp from previous posts. I have a few things to share about that later.

Ultimately, experimentation with a few gear additions took me back to looking at the head where I last left off. With the same shielding and same phono cartridge.

It's nice to take a break from things, and look at them again in a different light. I sure am glad I'm not on a time line...

In this round of testing, I need to go over the gear.
I'm using Protools to generate pink noise, and the Noise source is mono. I'm feeding the signal to the Phase Linear through a variable log attenuator so I can adjust levels. That's a new box I built especially for this use so I'll go over it another day.


I'm measuring voltage at the drivers for both right and left, to set power at 1W on each driver. I'm running the phono cartridge through the preamp that I modified. No RIAA EQ in any source or signal path. No loading at all on the torque tube (which will be required for a real test, and real results).

In this test I'm using two Ipads, and two Alesis IO/Docks for level changes and display, and a free Spectrum App. The free version has limited band display and the major annoyance of changing which channel it uses at random intervals. The software works great though so I'll buy it. Having two big spectrum displays without the need of a PC is very usefull.

The first shot is the output of the phono cartridge (using my modified RIAA-less preamp). It actually looks similar to early posts about this head with a big peak at 500hz, and another spike higher. Again, no load so not complete reality.

Important: The Grid is 5db. So it's pretty easy to see what is down by how much.

IMG_3505.JPG

Now notice there is a peak capture in the display. This is important, because I intentionally enabled that for other measurements we will see later. One thing to note, there is a big peak in low frequencies that is not real. That's when I moved the phono cartridge stylus into a position to make contact with the torque tube for the next measurement.

Important: The Grid is 5db.

The next two photos show the peaks captured from the previous photo, but in these shots the phono cartridge is in the same position but the stylus is not physically touching the torque tube. These photos are important. These photos tell us how much signal coming from the cartridge is actually magnetically induced and is not real signal that we can use for feedback. In both these photos I put "flags" at several locations for easy reference.

Here is the left channel with "flags" added. These same flags will be used for the noise floor shots later.
If you can't read the flags, let me type them. On the peak curve (top curve):

• 501hz, -26db.
  5,030hz, -44.
  13,181hz -63db.

The bottom has a reference at 498hz, -85db

So in the worst case, it looks like the left channel has about 20db of usable feedback signal at 16khz, with 1W at the drivers, and no load (not cutting a record). The numbers are higher at other frequencies, but that would be the absolute maximum we could obtain under these conditions.

IMG_3511.JPG

And here is the right channel shot with "flags":
The flags look similar:

• 504hz, -28db.
  5,010hz, -46.
  13,005hz -64db.

The bottom has a reference at 501hz, -85db

IMG_3512.JPG

One thing to note, it looks like there is a little 60 cycle noise on the left channel of the preamp. I may look at the shielding but it's low enough not to worry about. Especially when you see the next set of photos:

CONTINUED ON NEXT POST:

[EpicenterBryan]

And here are the left and right channel shots that still have the peaks and flags shown, but no audio signal at all. This is the noise floor using the phono cartridge, my modified preamp, and the drivers with no driving signal.

Left channel. Notice some 60hz noise. But way down.
And also the right channel below that. But super low noise floor.

IMG_3513.JPG

IMG_3514.JPG

This shows us that the magnetically induced signals are real. I also think there is enough signal that it makes this possible.

No time left to go into what was learned playing with the test bed (for tonight), but look for that in another post to tie that in as well.

Bryan

[markrob]

Hi,

Looks like really nice results! You should be in very good shape to close the loop. At this point, you can just try and close it and see how much feedback you can apply before the system oscillates or do more testing to characterize the open loop response.

If you go the latter route, you would need to run an open loop gain phase plot of the system and estimate the phase margin at your desired feedback level. That would allow you to develop a loop compensation network (lead-lad, notch filter, etc.) to reach your desired feedback level. Without access to a network analyzer, you have to do it manually with a two channel scope and signal generator (so you can measure phase shift). A pain, but you'll get good insight into the system.

Mark

[EpicenterBryan]

Hi,

Looks like really nice results! You should be in very good shape to close the loop. At this point, you can just try and close it and see how much feedback you can apply before the system oscillates or do more testing to characterize the open loop response.

If you go the latter route, you would need to run an open loop gain phase plot of the system and estimate the phase margin at your desired feedback level. That would allow you to develop a loop compensation network (lead-lad, notch filter, etc.) to reach your desired feedback level. Without access to a network analyzer, you have to do it manually with a two channel scope and signal generator (so you can measure phase shift). A pain, but you'll get good insight into the system.

Mark

Thanks Mark!
I bought an App for the iPads today for $14.95. It's a dual input O'scope that has cursors so I can make those measurements a bit faster than on my analog scope. For those interested it's called Oscilloscope By ONYX Apps.

I did find an app that totally looked like it did the job much easier and faster. It's called AudioTools by Studio Six Digital. The base App is $19.95, but the Transfer Function add on that looks like it does the gain and phase measurements directly is an additional $79.99.

So I opted to use an iPad scope and I'll take and calculate the data. I hate paying that much money for software without trying it out.

So, would you recommend acquiring data for both channels, and 20-20,000? Also, 10 measurements per decade?

Bryan

[markrob]

Hi,

That app looks nice. A steal for $79 if it works as advertised. I would do both channels. It would tell you how consistent your are. It might point out that a less aggressive amount of feedback is required if one channel has poorer response. It might be good to go an octave or so above 20Khz so you can see if there is a fast roll-off and rapid increase in phase shift. You might have to consider running your drive amp in a constant current mode to kill the L/R electrical time constant of your driver. Did you ever measure that?

Its hard to say at his point if you will be able to flatten all the way up to 20Khz, but it sure looks like you'll be able to do some significant feedback. Even if you only run to 10Khz, you can add some EQ outside the control loop to make the head flat. The big win will be the tight control in the midrange. You can estimate the high and low frequency closed loop response limits vs. feedback by making a quick Bode plot of you open loop response. Here is a nice tutorial:

http://www.ti.com/lit/ml/sloa077/sloa077.pdf

Looking forward to seeing how this works out.

Mark

[EpicenterBryan]

That app looks nice. A steal for $79 if it works as advertised. I would do both channels.

I agree. I sent them an E-mail today and have already received a response. My question was if I were to be able to get the two signals into the iodock adjusted to the same level at 1khz (my 0db reference somewhere below full scale) I would like to get a +/- db plot and a phase plot in degrees and if that was possible.

Here is what they said:
If you're using transfer function it will subtract the two signals directly, and produce a graph of the magnitude and phase difference between the signals. So if you can get the levels into the iodock it should work fine.

That is great news. And I have two of these setups so I can do it automatically for Left and Right at the same time. That brings up another possibility since the effort is now minimized. Should I do the same (right and left) at several drive levels? 1/2W, 1W, 1.5W. And should I think about rigging up way to load the head to simulate cutting force and at what kind of forces? I played with cutting forces early on and never really determined the ideal. As I recall 20g to 50g seamed to work OK with no heater and without heating the disc, and with using old blanks. I still don't have the heater set up.

So chime in everybody! If this app works out and the measurements can be automated it really makes taking lots of data possible.

It might be good to go an octave or so above 20Khz so you can see if there is a fast roll-off and rapid increase in phase shift. You might have to consider running your drive amp in a constant current mode to kill the L/R electrical time constant of your driver. Did you ever measure that?

No, I haven't made a measurement related to "L/R electrical time constant". Can you explain? As far as checking above 20khz, I could do that with my analog scope and a function generator. None of my gear goes over 48khz sample rate. Would one set of measurements at one drive level provide enough info?

Here is a nice tutorial:

Wow, that's good stuff Mark. I'll start looking into that.

So for the other guys, here is a video tutorial on the "Transfer Function" in app purchase from Studio Six. The main app is called "Audio Tools" - it costs $20, the Transfer Function add-on is $79.99. The iAudioInterface2 they mention is their box and is expensive ($499) But you can use other interfaces. I will be using the Alesis IO/Dock.

Bryan

[EpicenterBryan]

Wow. This is an interesting project.
First off, a big thanks to Mark for looking things over the last few days and setting me on the right path.

Groove scribe is looking really ugly and hacked these days. Here it is with the phono cartridge feedback mounted and the addition of a strain relief

IMG_3519.JPG

IMG_3518.JPG

I bought the transfer function app and generated transfer functions for both the right and left driver (one driven at a time). Let's look at those.
Ignore the remnants of another plot that I forgot to turn off near the top.

My intent was to adjust the graphs so that 1K was at zero, then scale them. Looks like I forgot to do that to the right channel, but you get the idea. They look similar but are not identical. In fact, the resonant frequencies are not the same.

LEFT ONLY:

Left Only Transfer Function.JPG

RIGHT ONLY:

Right Only Transfer Function.JPG

CONTINUED ON NEXT POST

[EpicenterBryan]

I really wanted to see how much feedback could be applied before it went crazy. To make it easier, I'm just going to post shots for the right channel. The left looks similar and responded in a similar way too. I have also edited out the phase plot. Since I needed an inverted feedback signal to sum, I needed to swap wires on the phono cartridge. But the phase plot needs a non-inverted signal or the plot makes no sense.

This first shot shows what happens right before the system goes unstable. Notice the big jump at those two frequencies. But also notice the center flatening out.

Right Channel With Feedback Applied - ignore phase.JPG

Next up, Mark suggested that I put a parametric EQ in the feedback loop, and see if I could bump those 2 frequencies down (around 4.5K and 13K). As it turned out both of my parametric EQ's are digital so putting one in the loop added an unknown conversion delay that messed things up so I swapped it for a 1/3 band graphic. What I ended up doing was cutting everything above 6.2Khz fully (-12db). This alowed me to apply even more feedback over a wider range before problems again in the 13K range. Now it's interesting to note that when the 13K peak started increasing it seamed to trigger 4.5K to also go whacko.

Right FB applied Cutoff 6.2K.JPG

Next up, I added an EQ outside the feedback loop to boost the lows 12db, Cut 5K and 13K (-12db) and boost the rest of the highs (+12) then adjusted the feedback level again...

Right PRE EQ FB Rolloff 6.2K.JPG

+/- 2.5db from 60hz to 4.4Khz or so is a good start. But until I can get more usable range (higher frequency) I'm not cutting any more test discs that's for sure! I know what this is going to sound like.

I think it's time for me to step back and see if the resonances can be identified and addressed. I have a few ideas I want to try out on my driver test bed. The one thing that has me really excited about this is that I can now apply a load at the stylus and the curve retains it's shape for the most part with a wide variation in pressure. And it also tracks as the drive level changes (within reason). That's been the killer on this project when doing test cuts. The curve was always different depending on cutting force, and drive level.

Bryan

[EpicenterBryan]

I had an interesting night last night, and today made a rather important discovery.

I was playing with the driver test bed, looking into some possible damping and other things I'll share later. But as part of digging into the phase issues in the high frequency I decided to go back and check my modified preamp. When I did, there was a large phase discrepancy. So I ran a spice simulation on the preamp including my modifications and the phase response was flat. I decided to step back and do a loop back test on the IO Dock and the iPad. I connected both outputs to both inputs, ran the delay finder and applied it, and got an identically looking and confusing phase plot as shown below:

IMG_0040.JPG

I ran the same test on my other IO Dock and got the same results. I bought a camera connection kit and connected the iPad (without dock) to a powered USB hub, then to a USB sound interface (Alesis IO2), did the same cable routing and got the same results. Then I contacted the App developer who was very responsive. After a few E-mails back and forth we figured it out and I want to share his response:

Probably there is a sample or 2 offset from the L& R channels on the output. You want to take just one output channel and Y it into both inputs, otherwise they are actually measuring different things. 180 degrees of phase shift at 20kHz would mean half a 20kHz wavelength delay on one side which is not much. 0.025ms if I am doing the math right. And one sample at 44.1 is 0.022 so I think that is what you are measuring.

So you are doing a transfer function of the outputs which are off by a sample.

Once I checked again using only one output and a Y cable I got a flat phase plot from the IO Dock and the IO2 interface. So I'm pretty sure I was using both outputs in the feedback testing; one to make the measurement, and the other to route to the summer. That means there was a big phase shift in the HF which in essence flipped polarity of the the high frequency in the feedback loop. This means that below 4K or so, the feedback would subtract, but above that it would add making the peaks even worse! I am unsure if there would have been a reason to have used 2 outputs for the normal response plots, but I will check them again to be safe. I'll only post them if they are different.

The take away is being 1 sample off can mess up the phase and your feedback at high frequencies! Wow.

Bryan

[opcode66]

Remember all of the discussion about DSP/DAC/ADC feedback. Yeah, this sort of confirms what I was saying there. One sample. The latency of any DSP/DAC/ADC will likely be at least that. Even if implemented without using a computer and OS.

[markrob]

And what will the phase shift be at 20 Khz if the sample rate is set at 200 Khz with a 1 or 2 sample delay???

[opcode66]

The part that crushes your argument is the processing time for the specialized dsp code.

[Bahndahn]

Processing may occur in the GHz range.

[opcode66]

I understand that. I've been writing assembly for pic chips since 2006. No handled environment whatsoever.

How long is the routine to handle resonance? Each instruction takes a minimum of 2 clock cycles, sometimes more. Those are assembly instructions. If we are talking about writing code in C and compiling, then it doesn't take long before simple routines start taking up a lot of those clock cycles to handle state and memory, etc. Especially if you are pulling in other libraries written by other developers with unknown efficiency. My guess is you would likely be doing some FFT here. So, what library will you be using for that? What latency and memory contstraints does it add? How long does it take for the cpu to read from the adc and write to the dac? If running under Arduino, there are other subroutines going on stealing processing time. So, given all that, what is the latency really? It doesn't take much for high frequency data to make this go awry. It matters little, minor fluctuations due to latency, when controlling a motor for real world scenarios. It is not the same when you apply these principles here with the expectations of perfectly flat audio response.