Continued sidetrack discussion from "Tiniest Portable amp I can build" - nikongod microtransformer-based impedance step-down box

Jun 5, 2011 at 11:34 PM Post #31 of 102
Lets Talk Measurements!
 
DISCLAIMER #1: These were taken in a less than scientific manner using an old USB SoundBlaster I almost forgot I had. The characteristics of the measuring device are less than ideal, particularly in the manner in which I had to take the tests (see below). So, take them for what they are worth. These will mainly serve as a RELATIVE comparison. 
 
DISCLAIMER #2: The output on the transformer-based boxes is down ~15db from the input. Because of this, I was unable to use the Line Out/Line In method to capture the data (not enough gain to register a clean measurement). So, these were taken using Line Out/Mic In, which, has a bit more gain to work with, but, is far less accurate (and is mono of course). So, do not be surprised when you see the SoundBlaster loopback results used for reference. Again, please remember the point is a RELATIVE comparison.
 
DISCLAIMER #3: The attenuation of the transformer boxes is the entire point of this little exercise. The sample levels for those devices were 100% output with the Mic input at ~43%. In contrast, that amount of signal completely swamps a standard loopback measurement made with a cable from Line Out to Mic In. The loopback reference had about 20% output with the Mic input at ~20%. So, a lot of signal was thrown away on the loopback reference vice the transformer-based measurements that enjoyed a completely unattenuated signal.
 
Nevertheless, here are the results. Again, the reference comparison is a loopback cable through the SoundBlaster card, the second is the original nikongod version of the circuit and the third is the jcx variation of the circuit:
 

 
Frequency Response (swept):
 

 
THD:
 

 
IMD + Noise
 

 
Dynamic Range:
 

 
Noise Level:
 

 
So, as you can see, the extra signal coming through is put to good use by the transformers. On the jcx Variation, dynamic range is up almost 6db and noise is down almost 6db. Distortion is held in check and the frequency response, save for a bump in the low end matches the loopback reference. Not too shabby. Good stuff.
 
Let the debate continue.... 
wink_face.gif

 
Jun 6, 2011 at 12:10 AM Post #32 of 102
Sweet work, thanks for measuring & posting. 
 
I'm going to stand behind my previously posted FR results, but still awesome results.Its nice to see that the THD & IMD are so low. 
 
Jun 6, 2011 at 7:23 AM Post #33 of 102
Your welcome!

Your FR measures are probably spot on for the technique and output levels for the measure. Certainly I would expect a decent scope to be far more accurate.

I think the issues with the high end of the frequency response are mainly a limitation of the SoundBlaster. Even using the line in/line out loopback method, the high end response of that card begins to roll off starting around ~15/16k or so and peters down to a -3db point around 19k. The microphone loop method is, as you can see, even worse.

I should probably pickup a more capable audio device if we really want to get something conclusive. In any case, I think this gets the point across.
 
Jun 6, 2011 at 6:24 PM Post #34 of 102
Very nice work Jake. RMAA is hardly what I would call "easy" to figure out or work with, but like you say, can be useful for some kind of visual comparison that you can share with others.
 
I have noticed that my Windows 7 laptop seems to like the latest 6.0 version of RMAA, after trying both 6.0 and 5.5 recently. On my setup(Lenovo T500 laptop and Emu 0404usb), I always seem to get a somewhat flatter frequency response when I have the windows microphone set to about +4 or +5.6dB. You can change the readout in windows by right clicking on the microphone slider button, and then selecting either dB or percentage.
 
As you have noticed, sometimes when testing you will find that it's hard to get acceptable levels to begin an RMAA test. Sometimes.
 
 
 
Interesting how the THD and IMD+noise lines are so smooth for the Nikongod box. 
 
Jun 6, 2011 at 7:03 PM Post #35 of 102
As you have noticed, sometimes when testing you will find that it's hard to get acceptable levels to begin an RMAA test. Sometimes.
 
 
 
Interesting how the THD and IMD+noise lines are so smooth for the Nikongod box. 


Sometimes? How about ALL times. :D

I used the latest version available (6.2.3). Run using XP hosted on a VMWare Fusion virtual machine. I found that setting the output to the highest level that did not clip and then optimizing the input setting worked best for consistent results. Even then, lot's of factors played into the measurements. Not least of which is the overall noise floor and stray RF interference induced by the test cables and location of the test itself. The cased up transformers were far easier to work with than the "loose" versions. I need to box up the original and give it another whirl.

Overall, I was impressed with how well the original version measured and pretty much tracked the cable loopback. Certainly a compelling argument for the simpler-is-better approach. While the jcx version measures better (and sounds better overall to these ears) the original version is quite impressive. Especially considering the cost of the parts.
 
Jun 6, 2011 at 8:23 PM Post #36 of 102
the smooth sloping  "skirts" on 2 of the plots are evidence that windowing setting must have changed - for a fair comparison all of the plots should show the same low/flat noise floor - I agree that you can't always tell what's going on in RMAA it can take a while to figure out levels, settings
 
 
 
Jun 6, 2011 at 11:40 PM Post #37 of 102
I agree, things do look a bit off with those distortion measures.
 
I went ahead and took another set of measures. This time, I used 24bit/48khz sampling as well as setting Master Volume at 94% and the Mic Input at 50% (0db supposedly). I then varied the Wave out level to set the proper recording level. I got these suggestions from the various guides on the Rightmark site and forum.
 
I have to think about these a bit more, but, they are most likely far more fair due to the fixed Mic input level used for all three samples. Again, a large amount of signal has to be attenuated with the loopback measurement.
 

 
Freq:
 

 
THD:
 

 
IMD:
 

 
Dynamic Range:
 

 
Noise:
 

 
These show a more consistent measure between the original Nikongod version and jcx variation. The main difference is the distortion characteristic which is more in line with what jcx stated would be the main benefit of the additional transformers.
 
I can certainly see how RMAA can me manipulated/misinterpreted. It is a fickle beast at best.
 
 
 
Jun 7, 2011 at 12:18 AM Post #38 of 102
Did you load the output of the soundcard and transformers to simulate the loads both see in real life? If not, maybe use a low(ish) impedance attenuator of 2.4K&240ohms to attenuate? Im not always such a fan of digital attenuation. 
 
Jun 7, 2011 at 7:30 AM Post #40 of 102
Did you load the output of the soundcard and transformers to simulate the loads both see in real life? If not, maybe use a low(ish) impedance attenuator of 2.4K&240ohms to attenuate? Im not always such a fan of digital attenuation. 


No, I did not. I can probably give that a go. Have to see what's in the parts bin. If i do not have enough parts, I can always put an Alps into the equation. I know I have a spare 50k in the bin. Note this would only be done for the loopback measure as the transformer measures are already near max output.

I think the main issue with displaying the data is that the transformer measures are taken at one setting and the loopback at another (output is attenuated digitally for the loopback). So, the windowing of the results when superimposed introduces display artifacts, just like jcx commented.

I will play with it some more when I get a chance.
 
Jun 7, 2011 at 10:01 AM Post #41 of 102

Users who are viewing this thread

Back
Top