First, what is NAS? According to Wikipedia, NAS is “a file-level (as opposed to block-level) computer data storage server connected to a computer network providing data access to a heterogeneous group of clients”. OK, so it’s a server that provides files to a computer over the network.

Another way to think of it is as a virtualized hard disk. An application running on the computer is probably not even be aware that the file storage it is accessing is elsewhere on the network.

By these definitions, I like NAS. What I don’t like is what “NAS” has come to mean when you buy a consumer-grade hardware/software “NAS”. Not to pick on any particular manufacturer but I’ll use Synology as I’ve owned a couple. I guess I have four issues with them:

• Cost/performance. For the money, these devices come with pretty low-performing CPUs and low amount of RAM. A “powerful” CPU in this world is a low end Celeron.
• Bloat. The original notion of “NAS” has been bloated to the point where these things have their own desktop and include their own package manager(s) that you can use to install all kinds of stuff… It looks great until you install some things and find out it doesn’t work that well or is hard to configure.
• RAID/poor scalability. I suspect the traditional RAID in the typical 4-drive consumer NAS has had its day. When you run out of space, you have to buy a set of new higher-capacity drives and… throw the old ones away.(*1) If a drive fails, the actual process of rebuilding the array is the most likely time that a drive will fail! (And if it does, your data is toast.)
• Reliability/single point of failure. Doesn’t matter how many drives you have, you still have only one power supply (remember: consumer-grade NAS). If that fails, no data access. That’s how my last Synology went, and why I won’t ever buy another.

(*1) Admittedly Synology did have an improved solution to this, their Synology Hybrid RAID. But you still have to replace at least two of the drives when you run out of space.

So what’s a better way – directly attached storage? No. I like networked storage. But I think these days, we need to think of networked services, of which storage is just one part. Point by point:

• Cost/performance. Almost anything is better. But for storage, an ODroid HC1, HC2 or H2 has loads more bang/buck, and can easily be set up as a dedicated file server. (There will be loads of other examples, these are just the ones I use/am familiar with.)
• Bloat. Start with a basic Linux server OS and add just the services you need. Don’t try and shoehorn all your networked services onto that poor little CPU in a NAS.
• RAID/poor scalability. 1. Use realtime sync to mirror your files across multiple machines. If one goes down, your files are all there on another machine(s). 2. Have another storage node on the network and write to it daily with a snapshot type of backup system. 3. Use a distributed file system like MooseFS, one or two drives per node. If you need more storage, add drives instead of replacing them!
• Reliability/single point of failure. One drive per node gives maximum power supply redundancy!

Not long after I purchased my ADI-2 Pro, RME released the “DAC only” version. It felt it worth a brief comparison. Here is the block diagram of the ADI-2 DAC:

As you can see by comparing to the block diagram of the ADI-2 Pro, the ADI-2 DAC is a much simpler device! Compared to the ADI-2 Pro, it is missing:

• Analog inputs
• The second pair of analog outputs and associated processing (i.e. Phones 3/4)
• Balanced headphone drive
• Digital outputs
• AES/EBU digital input
• Multichannel USB
• Rack-mounting holes

However, it gains:

• A dedicated IEM amplifier
• A remote control
• Incremental improvements in the internal clocking and analog output stage (these are now, I believe, in the currently produced ADI-2 Pro, which has received a name change to “ADI-2 Pro FS“.)
• Bass/treble and loudness compensation runs at 352.8 and 384 kHz sample rates
• An “auto-dark” mode
• Audiophile feet

There are some things that are just different:

• The unbalanced 1/4″ TS output jacks have been replaced with RCA jacks.
• The hardware reference levels are lower and uniformly spaced.
• The hardware reference levels on the unbalanced outputs are 6 dB lower than the balanced.
• The front panel is black instead of silver.

With the recent acquisition of the remarkable RME ADI-2 Pro, it’s time to revisit. This time, I’m looking for a simpler way to characterise performance. The early effort involved too many screenshots that were a pain to put into HifiZine articles. Plus, there wasn’t any way to say much about correlation to audible performance.

Introducing JoTT

The test signal I’m using now I call JoTT – short for John’s Torture Test (I had to call it something). I started by combining the CCIF and SMPTE test signals for IMD, and then modified it to make it most useful (for me). It has the following sine wave components:

The peak signal level is -1 dB FS. This avoids odd results that some DACs exhibit at 0 dB output.  (If 0 dB FS is actually required in the test signal, increase A to −6 dB FS or 50%.)

To generate JoTT (and plot the results), I am using the Electroacoustics Toolbox. Here is the spectrum of the generated digital signal (at 96 kHz sample rate):

Use as a test signal

For the sake of an example, I’ll use my Vioelectric HPAV200 with 24/96 coax DAC module (the older one). I’m driving it at the SPDIF input because the measurements of the amplifier alone are better than of the DAC. (For the purposes of explaining the test signal, it’s actually more helpful to use the worst-measuring input.) The output is taken from the left channel of the headphone jack.

I set signal levels so that 0 dB in the graphs is 1V RMS. First, here is the noise floor. There’s a very low amount of mains noise at 50 Hz and multiples of it. Also, there is some high frequency noise – I don’t think this is in the V200 itself, but is coming through ground somehow as I’ve just noticed that it appears in other measurements where I used a single-ended connection.

Here is component A (110 Hz) turned on. The cursors identify its harmonics:

Note: the levels of the harmonics is almost identical if I play just a 1 kHz sine wave at -8 dB. (If I play the 1 kHz sine wave at -1 dB instead, the levels and distribution of harmonics changes. However, you can’t have everything.)

If I now turn on component B (3 kHz), there is a small “forest” of intermodulation products around it. There is also a little thicket at 6 kHz, its second harmonic:

Now I’ll turn off A and B and turn on components C and D (9 and 10 kHz). You can see the intermodulation products at 1, 8 and 11 kHz in particular:

Now I turn on B, and you can see additional intermodulation products at 6, 7, 12 and 13 khz.

Finally, I turn on A again, and you see the “grass” of intermodulation products from around 1 kHz and above. This is the all-in-one graph, which I can use to point to both THD and to IMD in a way that clearly shows when a unit has issues – the worse the IMD, the taller and thicker the “grass” becomes.

Summary. From the single graph just above, we can read:

• The levels of the harmonic distortion components (multiples of 110 Hz).
• The level of 110 Hz sidebands around 3 kHz, and also 9 and 10 kHz.
• The level of the main IMD difference sidebands at 1, 7, 8, 12, 13 kHz.
• The thickness and height of the “grass” i.e. higher-order IMD components.
• Noise level from the mains supply.

Don’t get me wrong – this is good performance! If it “looks bad” that’s because the test is intended to show up problems. Hence the name “torture test.” I have others much worse, which I will may post at a later time…

JoTT is also useful for a quick crosstalk check, because of the multiple tones spread across the spectrum. I connected the right channel from the headphone amplifier and plotted its spectrum, while the signal generator was running into the left channel. You can see that there is some crosstalk, but it’s fairly low:

Reference loopback

Wait! How do I know that I’m not measuring the performance of my audio interface (the RME ADI-2 Pro, in this case)? Well, here is the graph when the analog output of the ADI-2 Pro is connected to its input:

Note: the DAC in the ADI-2 Pro is not being run at -1 dB FS. If I did that, there would be more distortion (although still very low). So this is not an “apples to apples” comparison of the ADI-2 Pro’s DAC with the Vioelectric DAC. Because that’s not what I’m trying to do here. The point is that the ADC of the ADI-2 Pro is being fed the same level of signal, and as shown in the above graph, the amount of distortion it introduces is insignificant (no more than shown, as the plot includes DAC and ADC distortion).

And here is the crosstalk for the ADI-2 Pro loopback:

Summary

What can I say, the ADI-2 Pro is incredible. I can now measure stuff I never could before. I will post some examples of JoTT measurements of other DACs soon, and hopefully the JoTT will appear in future HifiZine articles. If you have any suggestions or comments, please fill in the box below.

Pretty killer for the price, with a Dirac Live enabled version coming out as well.

Admittedly, the cabling on tiny boxes like these is not the most elegant…

The articles on the streamer start here.

Nothing yet on the 2×4 HD. Suggestions?

Note: for those unfamiliar with the LH Labs crowd-funding campaigns, the 1G USB cable is not a standard inclusion with the Pulse. It is included only for backers of the first campaign. If you are interested in buying a Pulse now, you will need to purchase the 1G cable separately if you want one.

Watch this video on YouTube.

As of this post, the Geek Pulse is still available on pre-order at a substantial discount off the retail price in LH Labs’ “Forever Funding” campaign. The Forever Funding campaign ends on December 27th.

I verified in Appendix A of my article The miniDSP nanoAVR – a Case Study that, while there is a measurable difference between the two orientations (pointed at the speaker with the zero-degree calibration file, vs pointed at the ceiling with the 90-degree calibration file), it is small and unlikely to have any effect on equalization.

One of the advantages of purchasing a microphone from Cross-Spectrum Labs is that you get a 90 degree calibration file included. However, you may already have a good calibrated microphone and not particularly feel like buying another. (I understand!) So, this article contains instructions on how to make a 90-degree calibration file. Your prerequisite, of course, is that your microphone already has a 0-degree calibration file. You also need to be able to use Room EQ Wizard (REW).

Background

When you make a measurement with a calibrated microphone, you are not changing anything that the microphone does. All that happens is that the software alters the measured frequency response in accordance with the calibration file. The math, in fact, is simple:

• Actual response = Measured response – calibration file response

The actual response is the “real” frequency response at a point in space (the tip of the microphone capsule). The measured response is what the measurement program will measure at that point in space without using a cal file. And the calibration file response is the magnitude column in the cal file (we’ll see an example below).

Let’s simplify the names:

• A = M – C

and rearrange:

• C = M – A

In other words, to get the 90-degree calibration file response, we only need take the measured response with the mic at 90 degrees, then we subtract the actual response with the mic at 90 degrees. But hang on, aren’t we trying to pull ourselves up by our bootstraps? – the actual response is the reason we are trying to do this in the first place and we don’t have that yet!

The thing is, though, the actual response with the mic at 90 degrees (using the 90 degree cal file) is the same as the actual response with the mic at 0 degrees (using the 0 degree cal file). So we can get A by pointing the mic at the speaker and measuring it using the 0 degree cal file.

Clear as mud? Never mind, just follow the steps below carefully.

Steps

1. Position the microphone pointed directly at a speaker and fairly close (to minimize room reflections). I used a small fullrange driver with the microphone positioned 20 cm away.
2. Load the 0 degree calibration file and take a measurement. Rename it to A (actual response).
3. Reposition the microphone so that it is pointed at 90 degrees. The tip of the mic must be in the same location as before.
4. Clear the calibration file and take a measurement. Rename it to M (measured response). [Note: leave the microphone where it is, as you will need it again later.]
5. Apply 1/3rd octave smoothing to both measurements.
6. View A, and go to File -> Export -> Measurement as text, and save the file A.txt.
7. Open A.txt in a text editor and locate the first line starting with a number greater than 1000. Delete everything before that line. Don’t leave a blank line at the start. Save A.txt.
8. View M, and go to File -> Export -> Measurement as text, and save the file M.txt.
9. Open M.txt and locate the first line starting with a number greater than 1000. Delete everything before that line. Don’t leave a blank line at the start. Save M.txt.
10. Open a spreadsheet (I used Microsoft Excel but others should work fine).
11. Position the cursor in cell A1 and import M.txt. Depending on the program the exact way you do this may vary, but you will typically use options that say it is a text file, with fields delimited by spaces or tabs. It should like a bit look this:
    
12. Position the cursor in cell D1 and import A.txt. Now it should like a bit look this:
    
13. In cell G1, enter the formula =B1-E1 :
     
14. Select all the cells from G1 down to G55, or whichever is the last row with numbers in it. Press Ctrl-D or locate the Fill Down command. The formula gets pasted in each cell.
15. Select all of Column G and do a Copy.
16. Select Column H and do a Paste Special -> Values only. This copies the result of the equation, without copying the equation itself.
17. In cell I1, enter the number 0.
18. Use Fill Down to paste the 0 all the way to the last row with numbers (I55 in my case). The result should look like this:
    
19. Select columns B through G and delete them. You should have three columns left, like this:
    
20. Use Save As.. to save the spreadsheet as a tab-delimited text file, named C.txt.
21. Make a copy of your calibration file, and name it something useful, like serial-90deg.txt.
22. Open serial-90deg.txt in a text editor and locate the first line starting with a number greater than 1000. Delete that line and everything after it.
23. Open C.txt, Select All and Copy its contents, then paste at the end of the open serial-90deg.txt file. Don’t leave a blank line. Save serial-90deg.txt.
24. You now need to verify that the calibration file works as intended. Back in REW, load serial-90deg.txt as a cal file (Preferences->Mic/Meter).
25. Run a measurement, and rename it to V.
26. In the Overlays pane, compare A and V. They should be very close. (If not, something went wrong.)
27. Optionally, you can use File -> Import Frequency Response to import serial-90deg.txt to see how it looks. (Set vertical limits to say +/- 10 dB.)

Example 1: CSL-calibrated Dayton EMM-6

I use the CSL-calibrated EMM-6 as my first example, because I already have a 90-degree cal file from Cross-Spectrum Labs for it. That way I can see how the cal file I came up with compares.

Here’s my A in blue and M in red (1/3rd octave smoothing):

Here’s my A again in blue and V in purple:

Good!!

Here is the comparison of the 90-degree cal file from Cross-Spectrum Labs in blue with the cal file that I just generated in purple:

Example 2: miniDSP UMIK-1

I don’t have a 90-degree cal file for the miniDSP UMIK-1, so I will generate one and verify it as described in steps 24-27.

Here’s A in blue and M in red (1/3rd octave smoothing):

Here’s A in blue and V in purple:

Here, out of curiosity, is the miniDSP 0 degree cal file in red and the generated 90 degree cal file in purple:

Additional notes

The procedure given above seems long and complicated, but it’s one of those things that takes longer to describe than to actually do. I tried to find a simple way of doing it, but this is the best I’ve found using readily-available tools (and no coding).

1. The procedure uses the existing data from the cal file below 1 kHz, and the new data above 1 khz. This is because the mic response due to orientation varies only at high frequencies. In addition, it’s hard (or impossible) to get repeatably accurate measurements at low frequencies because of ambient noise (traffic, planes, wind, footsteps, …)
2. A full range driver is best to use, as it avoids any directionality effect that might occur if positioning the microphone in front of a speaker with multiple drivers. If you don’t have a full-range driver, try:
   1. Moving the switchover frequency up. You may get a bit of a glitch though.
   2. Moving the mic further away (e.g. 50cm instead of 20 cm). This may give a slightly less accurate result due to room reflections but will probably be just fine.
3. I created column I for the phase data. However, most measurement programs will probably be fine without it.
4. The two times I’ve done this so far, there is a small glitch at 1 khz in the cal file. This could be avoided by matching A and M exactly at 1 kHz prior to exporting. However, the glitch I got was only around 0.1 dB so I decided it wasn’t worth the effort.
5. The impulse response of A and V does look a little different. I’m not sure at this point whether there’s ever a situation where this might matter.

Please let me know in comments below if you try this out and how you went. Thanks!

Postscript

You can, of course, come up with a method based on the above to calibrate one microphone based on another – but instead of rotating the mic 90 degrees, you measure using the second microphone. In order to get good low-frequency readings, the measurements should be done in two halves, as described here, and spliced together.

Note: these measurements are not intended to be authoritative or definitive. I took them to satisfy my own curiosity. If you know of measurements that are consistent with or contradict these measurements, please do post a link to them in a comment below.

Preliminaries: measurements in REW

I first tested three microphones with REW. These were all done in-room, approximately at a suitable listening position, since the iPad measurements will be simple 1/3rd-octave RTA. All microphones were pointed directly at the speaker and the appropriate (on-axis) calibration file used.

Firstly, a comparison of measurement sweeps using an Earthworks M30 vs my CSL-calibrated Dayton EMM-6 microphone:

REW Sweep, Earthworks M30 (red) vs CSL-calibrated EMM-6 (blue), 1/6th octave smoothing

The  two mics are very consistent across the range, with the EMM-6 reading slightly lower from 6 to 17 kHz. I’m not 100% convinced that the M30 is quite on the mark, for reasons I won’t go into here, but regardless, pretty close.

Next, a measurement sweep comparing the M30 to a miniDSP UMIK-1:

REW Sweep, Earthworks M30 (red) vs miniDSP UMIK-1 (green), 1/6th octave smoothing

The UMIK-1 reads slightly higher than the M30 below 50 Hz and above 8 kHz.

Now, I mentioned above that the iPad measurements will be an RTA (real time analysis), not a measurement sweep. Here’s how the REW RTA function (1/3rd octave) compared to the sweep, both with the UMIK-1:

Measurement sweep in REW (green) vs RTA in REW (red)

They’re pretty close, with the RTA drooping slightly at the top and bottom (the same effect was observed with the other two microphones). Note that the RTA does tend to bounce around a little, so some variation at individual 1/3 octave bands should be ignored.

iPad measurements

The measurements on the iPad were done using the RTA function of AudioTools. To compare measurements, the RTA was stopped and saved as a file, which was then moved to my laptop and imported into REW. The actual display in AudioTools is a bargraph, but they come out as smooth curves in REW. Good enough to compare though.

I was unable to use the EMM-6 and the M30 with the iPad as my MOTU Microbook II USB audio interface doesn’t work with the iPad. Bummer. I guess it saved me some time. So I used the UMIK-1 (with the Apple Lightning-USB adapter), a Dayton iMM-6, and the internal iPad mic. The UMIK-1 and iMM-6 had their respective calibration files loaded, while the internal mic used the AudioTools default (which apparently didn’t do much, see below).

First off, here’s the RTA measurement using the UMIK-1 taken on REW, compared with the RTA measurement using the UMIK-1 on the iPad:

RTA from REW (red) vs RTA from AudioTools (purple), both using UMIK-1

Pretty close, although there seems to be some drop-off in the AudioTools measurement below 90 Hz and I don’t think I’d trust the two lowest bands (i.e. below 30 Hz).

How does the AudioTools measurement compared in absolute terms? In other words, let’s assume that the sweep is the most accurate measurement. Here’s the sweep taken in REW using the UMIK-1, compared to the RTA measurement in AudioTools:

Measurement sweep in REW (green) vs RTA in AudioTools (purple), both using UMIK-1

Oooo… the AudioTools RTA reads about 3 dB lower from 70 Hz downward. Above 90 Hz, it’s as close as you could want. I can’t really explain this – if I can get the other mics working with the iPad at some point, I’ll see if I can verify whether this effect is consistent or just an anomalous measurement.

How does the internal mic compare? Here is the RTA from AudioTools using the UMIK-1 compared to the RTA using the internal mic:

Audiotools RTA using UMIK-1 (purple) vs the internal iPad mic (green)

As you can see, a fair bit of a drop below 200 Hz and above 6 kHz.

The little Dayton iMM-6, though, comes with a calibration file. Here it is compared to the UMIK-1:

Audiotools RTA using UMIK-1 (purple) vs the Dayton iMM-6 (blue)

It wobbles around a bit but is generally a decent match, with the exception of the 100-400 Hz range where it varies more than I’d be comfortable with for serious measurements or EQ. (I repeated the measurement and got a similar result.)

Cal file conundrum

After I took the RTA with the iMM-6, it occurred to me to check the calibration file. Loaded into REW, it looks like this:

Dayton iMM-6 calibration file

See all those wobbles above 4 kHz? They really shouldn’t be there. Dayton (or rather, their supplier) are doing something funny with the calibration. Still, for 1/3 octave RTAs, it makes no practical difference, but I wouldn’t use these cal files for measurement sweeps without smoothing. Here’s the Dayton cal file for my EMM-6 compared to the cal file from Cross-Spectrum Labs:

EMM-6 calibration files: from Dayton (green) and from Cross-Spectrum Labs(blue)

(In case it’s not obvious, the Dayton cal file simply can’t be correct.)

Conclusions/Recommendations

Based on the above measurements, here is what I think…

1. The internal microphone in the iPad isn’t really suitable for measurement work. It is probably possible to generate a calibration file that would make it work much better for this kind of RTA, at least above 40 Hz, but the default setting in AudioTools wasn’t it.
2. The little iMM-6 performed OK. If you’re just looking at the general trend and not trying to EQ something to the last dB, this is great value at $17. Or, it would be a great tool for learning the ropes without spending very much. If it weren’t for the funny response in the 100-400 Hz region, it would be a complete no-brainer at that price.
3. The external USB mic, the UMIK-1, performed consistently with the measurements from REW albeit with some droop below around 90 Hz. I can’t explain that but I’d assume that it’s the software not the mic itself. In either case, something to keep in mind if doing EQ. The Dayton UMM-6 would I assume be a good alternative, except for…
4. The Dayton cal files. If spending a bit more for a USB measurement mic (and the needed USB adapter for the iPad), then I would recommend getting the mic from Cross-Spectrum Labs. (Actually, I’d recommend that for the UMIK-1 as well as the UMM-6 – for a slight increase in cost over the stock mic you also get a 90-degree calibration file, which is more than a little handy if you’re trying to EQ an HT system i.e. with speakers all around you).
5. I don’t have a USB audio interface (i.e. with mic preamps) compatible with the iPad. For some reason I had assumed that the Microbook II would work, until I tried it and remembered that it does require a driver even on the Mac. Any recommendations on an iPad-compatible interface? Please post a link below if so.
6. iPad software is more limited than measurement programs running on a regular computer. It wins on convenience and portability but not on features or analysis capabilities. Nothing wrong with that, just be realistic.

Having now ordered and received some waveguides from Dave Pellegrene, I wanted to revisit the design of the Convertible. Based on the measurements of the SB29RDC in the waveguide and some other considerations, I’ve switched to a 6.5″ woofer, the Seas U18RNX/P. Here’s the concept design for a “convertible” monitor using the U18 and the Pellegrene waveguide:

Earlier, I had always assumed that the ported base of the “convertible” speaker would only add air volume (and a port). However, I recently had one of those “Aha!” moments, and with a little further investigation realized that a much better approach would be to add an extra woofer as well, thus turning it into a “2.5 way” ported floorstanding speaker. Here’s the concept diagram of that configuration:

Why the switch from 8″ to 6.5″? First, take a look at the SB29 response in the waveguide (0 30, 60, 90 degrees):

And here is Seas’ published U18 response:

If you look at the dispersion shown by the 60-degree curves (orange in the top graph, bottom curve in the lower graph), the directivity of the woofer gradually increases above a few hundred Hz and then matches the waveguided tweeter around 2 kHz. It’s almost as if these two were designed for each other!

The other thing about the U18 is that it has a nice little bit of extra Xmax, at 6 mm (peak). In fact, it has almost the same volume displacement as the U22 that I wanted to use in the original proposal. The net result is that I think two U18’s, while a bit more expensive, are a more effective solution than a single U22, for this speaker. And why not use two 8’s? Because the ported box would be too large.

I’m quite excited by this design (at this concept stage anyway). Everything – size, output, plate amp power, etc – just all seems to fit together really well. But first, I’ll try to finish off the Mini Convertible series of articles, translating this same concept to the “mini” format.

To mount the waveguide, you remove the three screws from the tweeter faceplate, insert the provided studs into the tweeter, and use the provided nuts and washers to bolt on the waveguide.

With a very quick and rough measurement, with the tweeter held by hand about 15cm away from the microphone, here’s the tweeter with the standard faceplate, at 0, 30, 60 and 90 degrees (ignore vertical scale except for relative reference):

And here it is mounted in the waveguide:

So that looks very promising!

Mini Convertible at Bathurst

As you can see, I made the second prototype with the tweeter offset in the baffle, so I can measure the effect it has on baffle diffraction. (Haven’t done the measurements yet.)

Some of the attendees:

Attendees at Bathurst GtG 2013

Terry, by the way, has an awesome DEQX-based system, which I’ve been meaning to try doing some sort of write up on for HifiZine. One day.

Photo Credits: Russ Tunny.