spectral waveform analysis of capacitors
Transcription
spectral waveform analysis of capacitors
SPECTRAL WAVEFORM ANALYSIS OF CAPACITORS (do the expensive ones perform significantly different from the inexpensive ones?) (does break in really alter a capacitors performance favorably?) By Carl Richard © 2007 INTRODUCTION A study was conducted to compare the audible performance of three different price levels of the same value capacitor from three different manufacturers. A key objective of the study was to make an advance in the heretofore, mostly anecdotal opinions of capacitor performance via listening experience by loudspeaker enthusiasts. As a consequence, a secondary (but equally important) objective was to see if the sonic signature of each capacitor was significantly different from the others. A test fixture was designed and built which allowed simple switching from one capacitor to another and at the same time make .wav and .aif type file recordings for later analysis using a microphone placed in close proximity to a tweeter. Signal sources feeding the test cap wired in series with the tweeter included a simple 8 kHz sine wave from a function generator, pink noise and a brief music selection. Saved files were analyzed using Raven Pro, Ver. 1.3 Beta software obtained under license from the Ornithology Lab at Cornell University‟s Bioacoustics Research Program. Although its primary use is for analysis of bird sounds, I found it a useful tool that allowed visualization of the tweeter‟s sonic output. The software has numerous features including correlation calculation of spectral data from 2 or more similar data files. Additionally, energy density plots, and large, numerical output .txt files are also possible wherein one can import them into an Excel spreadsheet and plot frequency, time and power (dB) levels. Waveform analysis has been around for some time now. It‟s nothing new. However, I have not seen any reference to its use in evaluating loudspeaker passive crossover components as a more quantitative, less subjective supplement to traditional listening tests. Being able to visually see and closely review/magnify a digitized picture of the sound emanating from a transducer downstream from a capacitor and do quantitative analysis appealed to the researcher in me. THE CAPACITORS A cap value of 4.7 microfarad (uF) was selected due its fairly common use in conjunction with tweeters to set a 1st order high pass crossover point in the 4 kHz range. A total of 4 caps were purchased for the study: 1) Radio Shack (50V) NPE #272-998 manufactured by Nichicon. 2) Solen 400V metalized polypropylene film (MPPF) 3) 100V boutique audiophile (MPPF) cap made by Hovland – Musicap. 4) Another NPE (100V) purchased from Parts Express to replace the RS cap Below is a summary of the initial measurements* taken on each, brand new cap. DESCRIPTION PRICE PAID uF ESR (ohms)** Df RS Electrolytic About $1 4.14 1.2 9.4% Solen FastCap About $3 4.83 ≈0 0.2% Hovland Musicap $32 4.84 ≈0 0.4% Parts Express NPE #027-332 About $0.50 4.89 0.3 3.1% * - measurements taken with a digital Tenma 72-960 LCR meter ** - measurements taken with an analog MAT Electronics ESR tester The relatively high ESR of the RS cap led to an initial frequency response test to assess its output vs the others. Below are scans of response plots showing the RS and Parts Express NPE‟s. Bottom scale is frequency and the side scale is un-calibrated decibels. fig. 1 fig. 2 In figure 1 above the RS NPE cap is a few dB lower in response than the Solen & Hovland caps. Figure 2 shows all 3 caps (PE cap now replacing the RS cap) super-imposed on one another (lower curve). The other curves were done when successive caps were switched on resulting the expected leftward shift and higher output. With the equal response performance of the Parts Express NPE cap confirmed, I proceeded to run initial waveform recordings of the caps. TEST RIG DESCRIPTION Figures 3 & 4 below are photos of the test unit. Built mainly from scrap wood, it included 3 DPST switches wired in series with each of the capacitors shown in fig. 4. All 3 caps were wired in parallel from the input to the tweeter. You can also see the LPG 26T 8 ohm tweeter. The very small cap in fig. 4 to the upper left is the RS cap disconnected from the test circuit and replaced with the Parts Express NPE cap. The design allowed switching on any one or all of the caps - the latter option being used for extended break in of 30 hours with a pink noise signal. fig. 3 fig. 4 Note the huge difference in size between the 3 caps. Both the PE NPE cap and the Hovland are each rated at 100 volts and yet the Solen 400 Volt cap is sized in between. Not sure why the disparity. RAVEN PRO ANALYSIS Figure 5 below is a screen shot of a waveform (top) and spectrogram (bottom) of an 8 kHz sine wave. Here I have deliberately stretched out the time scale until each X scale hash mark is 1/1000th of a second. Within the spectrogram, color is proportional to power level (dB). Thus the lighter areas represent higher level. Note the almost white band in the 8 kHz region and a less intense light band at the first harmonic (about 16 kHz). When working with Raven Pro, passing the cursor over the spectrogram displays both the frequency and dB level wherever the cursor is located. The vertical “Y” scale in the waveform has special Raven units. One “U” is the smallest amount of sound the digital recording can represent, the resolution of the recording (or at least one bit). fig. 5 Below, figure 6 is a composite screen shot of an initial test with a pink noise signal and successive switching on of each test cap separately without stopping the recording process. The kU waveform scale shows some higher peaks in both the Solen and Hovland caps relative to the NPE cap. My own zoomed in visual assessment of the 3 spectrograms led to the conclusion that the overall „hot spot‟ level of the Solen and Hovland caps was ever so slightly greater than the NPE cap. Note the rapid fall off of hot spots below 4 kHz in all caps due to the expected filtering action of each cap. fig. 6 NPE SOLEN HOVLAND Figure 7 below is a „direct from disk‟ recording of pink noise. Here we see an overall equal brightness to the spectrogram because there is no cap or tweeter to affect the normal, uniform energy density across the frequency spectrum associated with pink noise. This particular recording cuts off at 16 kHz. Compare this with figure 6 and you see how both the cap's filtering effect and the tweeter‟s natural roll off affect the color spectrum. Also note in the figure below the upper wave form is not stretched out and the spiny peaks shown here and magnified at the top of figure 6 are representative of the random energy emitted by a pink noise signal; both positive and negative. fig. 7 At this point I felt the next logical step was to record some real music to see if that variable would show any key differences in performance. I chose the first nine notes of E. Power Biggs‟ solo organ recording of Bach‟s Toccata in D minor. Figure 8 below shows the Raven waveform and spectrogram of the NPE cap. As with the pink noise signal shown earlier, we see the most intense power level varying a bit around the 4 kHz range depending on the notes played. The series of less bright horizontal bands above the brightest one are harmonics of the fundamental. fig. 8 Parts Express NPE Toccata Further examining of the NPE, Solen and Hovland Toccata figures was inconclusive. I think I may have seen a few more colored areas in the Solen and Hovland figures. If true, this would indicate those caps may be passing a bit more information than the NPE. This particular characteristic is the most important one in making a judgment on which cap may be the „better‟ one. Readers of this document are welcome to zoom in on figures 8, 9 and 10 to see if they can see any differences. However, I am not sure of how the quality of these images will survive following transfer to a .pdf document and subsequent transmittal via the internet. Fig. 9 Solen Fast cap Toccata Fig. 10 Hovland Musicap Toccata RAVEN CORRELATION ANALYSIS The software allows quantitative analysis of waveform and spectral outputs as long as the recordings are made under similar conditions, which was the case here. The correlation process involves the software mathematically superimposing one output over another from a different file. Thus, I was able to compare the NPE with the Solen and Hovland caps. The assumption here is if correlation is poor, then the outputs (i.e., the caps) are not performing the same. If correlation is good, then there is no difference between caps – something akin to the null hypothesis used in statistical student‟s “t” tests for significant difference, but more subjective. What it doesn‟t tell you is whether one cap is “better” than another. I used the correlation analysis on sine wave and Toccata recordings. Pink noise files were too variable due to their inherent nature and thus correlations were poor across the board. Below are screen shots of correlation outputs of the sine wave and Toccata files after 30 hrs of pink noise signal break in at a measured level of 1 to 2 volts across the cap/tweeter system. Fig. 11 Correlation (top), NPE (middle) vs Solen (bottom) – 8 kHz sine wave input signal Since the input signals are quite steady and consistent with sine waves, the correlation line (µ) is also seen as quite consistent in figure 11 above. At zero time, the coefficient is about 0.94. Fig. 12 NPE vs Hovland. NPE (middle), Hovland (bottom) 8 kHz sine wave input signal In figure 12, at zero time on the correlation chart (top), the coefficient at zero time is 0.88. Fig. 13 Hovland vs Solen. Hovland (middle), Solen (bottom), 8 kHz sine wave input In figure 13 on the previous page the correlation coefficient at zero time is 0.94. The results from the previous three 8 kHz correlations raise the question: Why were the coefficients good between Solen and NPE, Solen and Hovland but less so between the Hovland and NPE? I don‟t have a confident answer. Perhaps those differences in R values are not that significant? Next, correlations were done using the Toccata recordings thru each cap. Figures 14, 15 & 16 below show those results. fig. 14 NPE vs Solen Toccata. fig. 15 NPE vs Hovland Toccata Fig. 16 Hovland vs Solen, Toccata The odd shaped correlation lines are indicative of the variation in the music‟s signal strength and frequency. However, the coefficient values shown in red in the top window of each figure represent the peak correlation between files when they are superimposed. In this case they appear to be essentially the same. In each figure, the non-zero time offset at peak correlation is equal to the time offset between the start of each recording being compared. To further analyze the Toccata results among the 3 broken in caps, Raven software was used to generate power vs frequency curves of the caps following the break in period. The next 3 figures show those curves. These are perhaps the most telling since they represent the total energy recorded under exactly the same conditions with the test signal passing thru each cap on separate occasions. I believe for this study, total energy transmitted thru a given cap is directly proportional to its level of sonic quality. In this case I believe total energy does not necessarily translate to simply more volume (louder music), but instead, transmittance of more detail. Fig. 17 Parts Express NPE energy vs frequency spectrum plot The numbers in the lower left hand corner of each plot are: max dB and total energy. The max dB values are within < 1 dB of eacher and the energy values are within 0.6 dB of each other. Meaningful? I doubt it. Fig. 18 Solen Fast cap energy vs frequency plot Fig. 19 Hovland energy vs frequency plot The above results were compared with the same musical piece recorded when the caps were new. Unfortunately, the recording level at the new stage was higher than the broken in stage and as such absolute, definitive results are not possible. Similar energy vs freq. plots were generated and all were almost exactly the same: peak dB=73.5, total energy=84.5 dB. Proportionally, the energy level with the caps as new was somewhat higher relative to the broken in levels. Below is a table summarizing the measurements taken on the 3 capacitors after 30 hours of break in. There doesn‟t appear to be any significant change in measured properties. DESCRIPTION uF ESR Df Parts Exp. NPE 4.881 0.2 3.1% Solen FastCap 4.848 ≈0 0.1% Hovland Musicap 4.833 ≈0 0.2% CONCLUSIONS There was essentially no significant difference between the three capacitors evaluated in new and reasonably broken in condition. Certainly none to justify the pricey boutique type caps. The 30 hours of pink noise breaking in did not appear to appreciably alter any of the caps performance. It is not known if extending the breaking in period will improve performance. The Radio Shack/Parts Express NPE cap experience suggests if one intends to replace a NPE with a NPE, some caution should be exercised in the brand/supplier selection process. The high ESR and dissipation factor of the RS cap places this cap in questionable status as a crossover component as claimed in their packaging. A second cap was purchased from RS and tested even worse than the first. Raven Pro waveform analysis software can be a useful tool for evaluation of passive crossover components. **************************** DISCUSSION In my mind there still seems to be one elusive characteristic I have yet to figure out; it‟s a way to measure „smoothness and/or harshness‟ of the sound. I don‟t think waveform and/or spectrograms are the right tool for characterizing these important sonic virtues of audio components. However, I could easily be mistaken here simply because I am a brand new user of waveform analysis tools. Once again, the basis for using these tools in this study was to see if any cap could deliver more information. Manufacturers and others have made claims for the high-end caps being more „revealing‟ with „broader soundstage‟ and so forth. The spectrograms would have shown this by simply revealing significantly more bits of sound, but they didn‟t. Based on this study I am gradually becoming convinced that the „harsh‟ sound from new, MPP caps which replaced old, NPE or paper/oil caps that have leaked or drifted in their properties may simply be more decibels passing thru it to the tweeter from the relatively lower ESR of the new cap. Secondly, IMHO, what we believe to be a smoothing of the sound over time is the scientific documented effect of psychoacoustic masking. It must be noted here that this study basically represents a sample size of one. Broad conclusions extending beyond the scope of this study relative to type, brand, break in, etc., are not recommended without further sampling and testing.