Part II: A MULTI-TONE INTERMODULATION METER CONSTRUCTION
BY ERNO BORBELY, Contributing Editor
IN PART I OF THIS SERIES of articles I described the most important sources of nonlinear distortion and some measurement methods for these nonlinearities (TAA 2/89). Most of the measurement methods are general purpose, measuring either static (1-kHz THD, SMPTE IM) or dynamic distortion (high frequency THD, CCIF IM). Some test methods have also been developed for detecting a particular type of dynamic distortion associated with the slew limit of the amplifier (DIM 30/100).
With these tests you cannot measure one or the other type of distortion exclusively, which, except in design work where you want to pinpoint a particular distortion mechanism, is probably not important. The more important characteristic of any measurement is whether it produces good correlation between the test and the auditory results. In this respect we know that certain types of tests have a better chance of producing good correlation than others.
1. Tests that show poor correlation rely on out-of-band harmonics (high frequency THD), or use, to a large ex tent, out-of-band harmonics to generate in-band IM products (DIM 30/100); or those where the auditory masking effect is high (SMPTE, DIM 30/100).
2. For tests that have a much greater chance of producing good correlation, input stimuli and IM products are all in-band, the input stimuli are far re moved in frequency from the IM products, thus avoiding masking, and feed back is working at its normal level to attenuate the IM products.' Unfortunately, even here we don't have absolute criteria that can be used to design good sounding audio equipment; more work is needed to establish such criteria.
Figure 1 is reproduced with permission--R.R. Cordell and the AES. (JAES, vol. 29, no. 9, September 1981, p. 583.) By all means, I do not advocate that tests in group 1 should be abandoned completely. On the contrary; for example, I use the high-frequency THD extensively for testing high frequency nonlinearities. All I am saying is more effort should be put into measurement methods that are likely to produce good correlation.
If we now agree on the general desirability of group 2 tests, we should look at those that can exercise the widest range of nonlinearities, including the high frequency ones. For high-frequency testing the rate of change is important.
The three IM tests mentioned in Part 1 have the following rate of change:
FIGURE 1: Block diagram of (a) MIM test signal source and (b) MIM product analyzer, as proposed by Cordell.
FIGURE 2: High-stability test oscillator for the multipurpose IM meter. Simpler configuration can be used if frequency stability is satisfactory.
10.05kHz: C1 =C2 =1nF 9kHz: C1 =C2 =1.18nF C1 and C2: dipped mica or COG (NPO) ceramic P1-3: 10-turn cermet (Resista RJ-5W or equiv.) D1-3: IN4148
CCIF IM: 0.1225V/ u-sec per volt peak Cordell MIM: 0.08179V/ u-sec per volt peak
IEC IM: 0.06267V/ u-sec per volt peak
The standard CCIF test's disadvantage is its inability to detect odd-order non linearities. Hence, theoretically, it is less suitable for testing modern, symmetrical audio amplifiers, where odd-order nonlinearities are dominating. How ever, our devices are far from 100% sym metrical in the world of practical audio, and the CCIF test will detect these asymmetrical nonlinearities. Oddly enough, it has proved a very good test for hard slewing related distortion because most amps tend to slew limit in one direction sooner than in the other.
The Cordell test corrects the deficiency of the standard CCIF test by adding a third tone and producing an odd order product of the type f3-f2-f1, close to the difference frequency.' The penalty for adding a third tone and changing one of the high-frequency tones to a lower one is a 33% decrease of the normalized rate of change of the composite signal (from 0.1225 to 0.08179V/ u-sec per volt peak). Therefore this test is only two-thirds as likely to exercise slew related distortion.
The sensitivity of all IM tests mentioned depends not only on the highest rate of change, but also on how many times the test is reaching this rate of change in a given time. Again, the CCIF has a distinct advantage: it peaks more often than the Cordell test signal. Consequently, measuring purely asymmetrical nonlinearities, the CCIF test shows a significantly higher sensitivity than does the Cordell. However, with purely odd-order linearities it shows nothing, whereas the Cordell shows all odd-order distortion products. Clearly, both tests have advantages and would be useful in a general-purpose IM meter.
The IEC IM test, with half the rate of change of the CCIF test, does not, in my opinion, meet our needs of testing modern audio amplifiers. However, it is a good, general-purpose IM test for detecting a wide range of even- and odd order nonlinearities, which is strongly recommended for testing band-limited systems such as tape recorders, speakers and transmission lines. As my primary interest is in wideband systems, assumed to be flat to 20kHz, I didn't consider the IEC test to be optimal for my instrument.
Cordell's Multi-tone
When I set out to design my IM meter, I first considered the Cordell test, soap plied to Mr. Cordell and the Audio Engineering Society (AES) to use his idea in a construction project for TAA. This permission was kindly granted and I made a prototype, which I have evaluated on a number of amplifiers (discrete and ICs), comparing it to 20kHz THD measurements. As the Cordell test uses 950Hz and 1,050Hz product frequencies, and one of the test frequencies is 20kHz, it was easy to add the standard CCIF test to the instrument by adding a 19kHz oscillator; the 1kHz difference frequency falls in the middle of the analyzer's band width. Later I introduced a different combination of frequencies to make the instrument even more useful, and I will present it in this final form.
Mr. Cordell's block diagram for the MIM test source and the MIM analyzer is reproduced in Fig. 1. The oscillators are single op amp types with an FET, producing 5V RMS output with approximately 0.01% THD. Because the test is independent of the oscillator THD, even more could be tolerated. However, low noise and very good frequency stability are of primary importance.
The first is necessary for a low measurement floor, the second to avoid drift, which might cause zero beat between The first is necessary for a low measurement floor, the second to avoid drift, which might cause zero beat between the product frequencies. For proper operation the product frequencies should not drift more than +20Hz. This can be divided into 50% each of initial adjustment and drift. Initial adjustment should thus be +2.5Hz for the 9kHz and 10.05kHz oscillators and + 5Hz for the 20kHz oscillator. The remaining +10Hz allowance for drift seems rather tough, but can actually be met without much difficulty if the same kind of high-stability tuning components are used in all three oscillators.
Because the lowest test frequency, 9kHz, is almost a decade above the product frequencies, only moderately sophisticated filtering is necessary in the product analyzer. Mr. Cordell uses two third order, 2kHz Butterworth low-pass filters, with 20dB gain after each to maintain good signal-to-noise ratio. This gives more than 70dB attenuation at 9kHz. Mr. Cordell also uses a fourth-order bandpass filter, with a 200Hz band width. This rejects a second, odd-order product at 2kHz, reduces the noise band width of the analyzer--thus improving the measurement floor--and reduces low-frequency interference, mainly from AC power line sources. Mr. Cordell's source-analyzer combination has a measurement floor of 0.0002% referred to a peak test signal of 1.41V (a composite, 1.41V peak signal is equivalent to 1V RMS sine wave).
The Signal Source
Although I am using different oscillators and filters, my multi-purpose IM meter is similar to that of Mr. Cordell's block-schematic in Fig. 1. The high stability oscillator (Fig. 2) used in the signal source was designed by Mr.
Kuroda and published in Linear Application Update (# AU-E005) by National Semiconductor. Although intended as an ultra-low distortion, 10kHz spot-frequency oscillator (and I am using it for that purpose in my THD measurements), it also features good frequency stability, so I made some extra boards for the IM meter, with some modifications to the original circuit. Mr. Kuroda has been using LM833s, but I had stability problems with these devices at 20kHz in my layout, so I changed to LF412CNs. This increases the noise, but the increase is practically unmeasurable both in the 20kHz THD test and in the IM tests. The 2SK170/2S 74 FETs must be from I_pss group BL for proper operation. I have also added regulators to each board, as I have done on all the IM meter boards.
The most critical components in the oscillator are the tuning elements: P1, C1 and C2. P1 should be a high stability (both electrically and mechanically), multi-turn cermet trimpot. The capacitors should be dipped mica or, preferably, COG (NPO) ceramic. Using COG ceramics and Resista cermet trim pots, my oscillators are still within a couple of hertz of the nominal frequencies after many months of use. As I mentioned before, the THD of the oscillators is not important, so if you have a design that satisfies the frequency stability and noise requirements, feel free to use it.
To implement the Cordell test I made three oscillators, tuned to 9,000Hz, 10,050Hz and 20kHz. The CCIF test, using 19kHz and 20kHz, could be implemented by switching the 9,000Hz oscillator off and changing the frequency of the 10,050Hz oscillator to 19kHz. For stability reasons, however, I have added a separate 19kHz oscillator so I don't have to switch tuning elements. I always run all four oscillators continuously and switch off the output of those not needed, eliminating warm up drift when switching frequencies.
This makes the instrument a bit more expensive, but also more practical because I can switch operating modes in seconds.
The four test oscillators are fed to a summing amplifier, which combines two or three of the tones, according to the mode of operation (CCIF or Cordell). Let's look at the signal levels at the different parts of the test signal source for the Cordell and the CCIF modes. To test ICs in the unity-gain follower connection to near clipping, you need a minimum 20V peak-to-peak (p-p) composite waveform. This is equivalent to a 7.1V RMS sine wave.
In the Cordell test the three sine waves have equal amplitude, so the peak amplitude of the composite waveform will be three times the peak amplitude of one component. Each signal must therefore be 6.66V peak or 2.36V RMS.
Now, if the oscillators were running at 2.36V RMS output, they could be combined in a unity-gain summing amplifier, with R1 + R2 = R (Fig. 3a).
(R1 and R2 are two resistors in series, see next paragraph.) However, since my oscillators are running at 6-7V RMS, the gain must be reduced to less than unity. I have adjusted the oscillators to 6.5V RMS, and selected R1 + R2 = 27.4k-Ohm, which gives a gain of 1/2.74 with R = 10k. This produces a composite test signal of just over 20V p-p.
Switched Modes
In the CCIF test the composite wave form will have twice the peak value of the two components. To have a 20V p-p amplitude at the output you need two 10V p-p signals, which is equivalent to 3.55V RMS. Again, the oscillators are running at 6.5V RMS, so you need an attenuation of 6.5/3.55 = 1.83 times in the summing amplifier (Fig. 3b).
This can be achieved by using R = 10k-ohm and R1 = 1.83 x R. Closest standard value is R1 = 18.2k-Ohm). In my signal source the switching between the Cordell and the CCIF mode occurs by shorting out part of the 27.4k-ohm resistor.
The complete summing amplifier is shown in Fig. 4. It has four inputs, where each input has a series resistor of 18.2 + 9.31 = 27.51k-ohm.
FIGURE 3: Signal levels and gain setting in the (a) Cordell and (b) CCIF
modes of operation.
FIGURE 4: Summing amplifier for multipurpose IM test signal source. Note
the floating output to avoid ground loops.
In the Cordell mode the whole 27.5kf} is in use; in the CCIF mode a switch shorts the 9.31k1 resistors. I made the layout for a four-pole DIL switch, which is accessible from the side of the box with a screwdriver. Unused inputs, or if you wish, unused oscillators, are disconnected by grounding the respective summing amplifier inputs. I used switches on the back of my unit, but if your box is big enough you can put all four switches on the front panel.
The composite output of the test signal source is adjusted by a 2.5k2 linear pot, which must be high quality to avoid intermodulation products of its own. I am using a conductive plastic type here. Note that the output connector (in my case a BNC type) is floating (insulated from the chassis) to avoid ground loops in the test setup.
IM Analyzer
In the block schematic of the product analyzer and associated circuitry (Fig. 5), the analyzer itself consists of an eighth-order, 2kHz Butterworth LP filter and a fourth-order Butterworth bandpass filter of the state variable type. My attempt to duplicate Mr. Cordell's arrangement with two third-order LP filters didn't work because it generated too much noise. The final arrangement described here evolved through a number of previous prototypes: the first used a sixth-order LP and second-order BP, the second used the present eighth-order LP, but still retained the second-order BP; I finally upgraded the BP to fourth-order, partly to get a higher attenuation of the un wanted, second odd-order product at 2kHz that I mentioned earlier. The combination of a sixth-order LP and the fourth-order BP is also a good solution and you can save a couple of ICs in the process.
For convenience I have divided the analyzer part into three boards, called F1, F2 and F3. F1 contains an input buffer, sections 1-4 of the eighth-order
------ FIGURE 5: Block schematic of IM analyzer.
LP, and a 20dB gain stage (Fig. 6). The second half of the LP is on the F2 board, containing sections 5-8 and one 20dB gain stage.
Let's now look at the calculation of the filter components for F1 and F2.
The normalized values of the capacitors for an eighth-order Butterworth LP are:
Cn1 = 1.18
Cn2 = 0.846 Cn3 = 2.05
Cn4 = 0.437
Cn5 = 0.965
Cn6 = 1.04
Cn7 = 4.84
Cn8 = 0.207
The actual capacitor values with equal value resistors are calculated from:
C = Cn (1/wo x R) (2)
Here you must experiment a bit to come up with practical values for the resistors and capacitors. The trick is to find a resistor value low enough for low noise and also using capacitors small enough so the op amp can drive them without distortion. A resistor (R) less than 10k is a reasonable compromise.
Some calculations with standard value caps show that R = 9.31k-ohm is a good choice. For a 2kHz filter, formula (2) gives:
C = Cn x 85475 x 10° (3)
Inserting the eight Cn values in formula (3), you get the following eight capacitor values for the filter:
C1 C2 C3
0.01 uF
7.23nF
0.0175 uF
First half of LP filter (F1) 8.24nF 8.88nF
0.0413 uF 1.77nF
CY C2 C3' C4'
Second half of LP filter (F2)
Eye.
C1 and C1' are standard values; the others require two capacitors in parallel. I designed the layout so two capacitors can be connected in parallel for all positions. They must be high quality film types (polystyrene or polypropylene) or dipped mica. Despite the popular belief that ceramic is no good, the COG (NPO) ceramic caps are highly suitable for filter use (and a lot of other things) in audio.
The input buffer on F1 is an NE5534, as are the two 20dB gain stages on F1 and F2. The filter sections use dual op amps; I tried the LF412 and the MC34082 and found both work satisfactorily. F1 and F2 use the same layout; the schematics are the same, except F2 doesn't have an input buffer.
FIGURE 6: First half of the eighth-order 2kHz Butterworth LP filter (f1).
Second half is the same, except no input buffer and different capacitor values
(f2).
I am using separate fixed 15V regulators on all boards: the LM340LAZ-15 and LM320LZ-15, improved versions of the 78L15 and 79L15. The 78/79 series may also be used on all boards, so I have indicated these on the schematics.
Recipes If you are looking for a measurement floor on the order of 0.0001%, you must attenuate the test signals, specifically 9kHz, by more than 120dB. This LP filter attenuates 9kHz by more than 100dB, so adding another 20dB or so is no big task. Indeed, a second-order state-variable BP filter with center frequency = 1kHz and BW = 200Hz is about 30dB down at 9kHz. Unfortunately, its attenuation at 2kHz where you have an unwanted IM product is only 17dB, which I consider too low.
I use two second-order state-variable filters in cascade (Fig. 7). This type of BP filter is described in detail in Audio IC Op-Amp Applications. Bandpass gain is given by:
BP gain = Q (R6/R7) (4) and Q = R4/R3 (5)
If you want a bandwidth of 200Hz with a center frequency (fc) of 1kHz, then Q = fc/BW = 5.
Calculate the necessary component values for the two BP filter sections by:
section 1: Gain = 1 (unity) Q=5
Assuming R4 = 10k, R3 can be calculated from (5): R3 = R4/Q = 2k.
Assuming R6 = 10k, you can calculate R7 for unity gain from (4):
R7 = (QxR6)/BP gain = With R4' = 10k, R3' = 2k). With R6' = 10k, R7' = 5k.
Center frequency is given by the formula (assuming R1 + P1 = R2 + P2 and C1 = C2):
50k.
fc = 1/[2x(R1 + P1)C1] (6) section 2: Gain 10 (20dB)
Selecting a standard value 0.01uF = capacitor, R1 + P1 = R2 + P2 = 15.9k-ohm.
Now, if you cascade two BP filters with fc equal to 1kHz, the response will not be flat at the two product frequencies: 950Hz and 1,050Hz. In fact, you should go through a rigorous calculation to find out how to ‘tune’ the two filters to get a band-pass ripple less than a given value. I have done this, but in the final analysis, I experimentally tuned the first section to 900Hz and the second one to 1,100Hz, which resulted in -0.2dB down on the two product frequencies, compared to 1kHz:
R1 + P1 = R2 + P2 = 17.68K-ohm for the first section and R1' + P1' = R2' + P2' = 14.46k-ohm for the second.
FIGURE 7: F3-two second-order state-variable BP filters connected in cascade
to form a fourth-order BP filter.
Tuning is done by the trimpots P1 =P2 = P1' = P2' = 2.5K-ohm.
Attenuation through the two sections (fourth-order BP filter) is equal to 29dB at 2kHz and 59dB at 9kHz. The gain of the two sections is approximately 6dB less than the calculated gain because of the staggered tuning. I have adjusted this by changing R7 = 22.1k-ohm, which brings it back to just over 10 times (20dB). I fine adjusted the total gain of 60dB through the whole analyzer by using the gain control in the 20dB gain stage on F2.
The analyzer is actually complete with the three filter boards F1, F2 and F3. All you need is a pot at the input to adjust the reference level, and you can measure IM (Fig. 5). In fact, I used the analyzer in that form until recently. The measurement procedure is simple: adjust the signal source output to produce the necessary voltage swing at the output of the amplifier under test, measure this level on a peak-to-peak reading instrument, adjust the input reference for the analyzer to 1.41V peak or 2.82V p-p, and read the output of the analyzer on an RMS calibrated, average responding millivoltmeter. The sensitivity of the analyzer, with gain adjusted to 60dB, is 0.0001% per mV as read on the millivoltmeter. The mini mum input voltage to the analyzer, when using 1.41V peak (or 2.82V p-p) is about 3V p-p.
FIGURE 8: Input buffer and peak-to-peak detector.
FIGURE 9: Input configuration when using the peak-to-peak detector. Note
input grounding and use of twisted pair wiring.
FIGURE 10: Dual +- 20V unregulated supply for IM meter.
Easy Signal Monitor
Now, the only time-consuming part in this is to read the output of the amplifier in volts peak-to-peak, and adjust the reference to 1.41V peak or 2.82V p-p. I have done this on a scope for many months, but it takes time and I have been looking for an easier way. My friend Dr. Kalman Molnar suggested including a peak or a peak-to-peak detector in the instrument, which I did (Fig. 8).
The circuit consists of a buffer and a textbook positive and negative peak detector that you can find on just about every op amp data sheet. ’ Monitoring the plus and minus outputs requires a symmetrical, floating DC voltmeter (DVM). Some types are not suitable because the ground side injects noise into the analyzer, which shows up as increased noise, raising the instrument measurement floor. If you have such a DVM, connect it between ground and positive, thus measuring peak instead of peak-to-peak (Fig. 9). When you then make your measurements, you simply multiply this reading by two to get the peak-to-peak value.
Because of the limited voltage swing capability of the peak-to-peak detector, I installed a voltage divider at the in put of the analyzer. I chose a 6dB per step attenuator, which allows you to monitor signals up to 160V peak-to peak. The impedance of the attenuator is 100k, which is connected in parallel to the normal input divider. The total input impedance of the analyzer with the peak-to-peak detector is: 100k in parallel with 10k = 9.1k.
Power Supply
Although you can get away with a single power supply for the signal source and the analyzer, I recommend running the signal source (oscillators and summing amplifier) and the analyzer from separate supplies. This will assure that you don't run into ground loop problems in your measurement setups. The dual power supply is shown in Fig. 10. I have indicated a single transformer with four 15V windings, but you can just as well, as I did, use two small (15VA) transformers with dual 15V windings. Although the power supply board has no regulator, I have fixed *15V regulators on all boards. However, if you prefer to have a central regulated supply, feel free to make your own design.
I have also tried the IM analyzer with a single supply feeding the signal source and the analyzer, and it worked okay in my measurements. All I had to do was to break the ground at the output of the amplifier under test and leave all other grounds connected.
Layouts and Guides
Figures 11-16 show the circuit board patterns and the stuffing guides for the IM meter. Each function is on a separate board for flexibility. However, there is no reason why you can't put the whole IM meter on two boards: one for the test signal source (oscillators, summing amplifier and power supply), and one for the analyzer (filters F1, F2 and F3, peak-to-peak detector and power supply). Also, if you believe you don't need some of the IM measurements, you could leave out those modules. For example, if you want only CCIF, you will need only two oscillators.
Figures 11a and 11b show the lay out/stuffing guide for the high-stability oscillators. My circuit board calls for several jumpers; surely you can improve the layout, if you wish. Figure 12a shows the layout for the summing amplifier. The DIL switch on the summing amplifier board (Fig. 12b) lets you short out part of the input resistor, which is necessary when you switch between the two-tone (CCIF) and the three-tone (Cordell) operation mode. If you put this switch (or switches) on the front panel, make wires to the switch very short.
The layout for filters F1 and F2 is the same; the stuffing guides (Figs. 13b and 13c) show the difference between them.
F1 uses a buffer stage and has a capacitor in series with the input. The capacitor prevents DC offset from the amplifier under test reaching and overloading the DC-coupled filters. The three test points -TP1, TP2 and TP3--offer an easy way to test the different stages on the F1 and F2 filter boards. All filter boards have separate signal and supply grounds. The signal ground is strapped from F1 to F2, then to F3 and is connected only to sup ply ground on F3 (Figs. 14a and 14b). This works fine in my unit, however, feel free to try other grounding schemes in your IM meter.
You will notice some unused real estate around the input buffer on the peak-to-peak detector board (Figs. 15a and 15b). I wanted to use this stage as a switched-gain input amplifier for the whole analyzer, but gave up the idea because of complicated switching arrangements. In the present circuit, I use it only as a unity gain follower which serves to isolate the analyzer from the peak-to-peak detectors.
The power supply board, shown in Figs. 16a and 16b, contains two independent, unregulated supplies. It de livers approximately +20V DC with 2 x 15V RMS on the transformer. All boards work with + 18V, so you have a couple of extra volts for taking care of variations in the AC supply voltage.
Metal Mechanics
Since the mechanical layout is not critical, I am not proposing any particular solution. My IM meter has under gone numerous mods (adding the fourth oscillator, the peak-to-peak detector and so on) so the layout, especially that of the front panel (Photos 1 and 2), is not optimum. I would use a bigger box (the present one is approximately 10 by 10 inches) so I could put the test signal source on the left side, the analyzer on the right and the power supply in the back. Most importantly, the oscillator on-off switches should go on the front panel, to make switching between the different modes of operation more convenient.
FIGURE 11b: Stuffing guide for high stability oscillator.
FIGURE 12b: Stuffing guide for summing amplifier
FIGURE 14b: Stuffing guide for F3.
FIGURE 13b: Stuffing guide for F1.
FIGURE 15b: Stuffing guide for peak-to-peak detector.
FIGURE 15a: Layout for peak-to-peak detector.
FIGURE 16a: Layout for power supply 5.
FIGURE 16b: Stuffing guide for power supply 5.
PHOTO 1: The front panel of the IM meter. Note the floating BNC connector on the test signal source output.
PHOTO 2: Detail of the front panel. Note the Elma switch with 1% metal film | resistors used as the input attenuator for the peak-to-peak detector. The potentiometer behind it is a 10k/ten-turns type, used as the input pot (marked CAL on the analyzer front).
PHOTO 3: One of the oscillator boards and the summing amplifier board. Note | the on-off switches for the oscillators on the back of the unit, and the gain switches on the summing amp board.
I have metal shields between the oscillator boards (Photo 3) to prevent interaction between the oscillators. I have also shielded the rest of the signal source (summing amplifier) from the analyzer (Photo 4), but I do not use shields between the analyzer boards. I used two small transformers in the unit because I didn't have room for a bigger one with 4 x 15V secondary. You could use a single transformer with a larger box or if you use fewer modules.
As I mentioned before, you can per form the Cordell and the CCIF tests with the IM meter I described. In fact, you can do a second CCIF test with this IM meter at no extra cost: using f1 = 9kHz and f2 = 10.05kHz you will get a difference tone at 1,050Hz. As you recall, the Cordell test uses the same frequencies to generate the difference product. Again, this test only detects even-order products caused by asymmetrical nonlinearities. Its rate of change is 0.0598V/usec per volt peak, which is similar to the rate of change of the IEC test. Thus, it has limited use for slewing related distortion, but can be useful in band-limited systems.
A New Three-Tone
Although I used this IM meter in this form for some time, recently I started to experiment with the frequencies involved. I believe a three-tone IM test should be closer to the CCIF test in terms of maximum rate of change than the Cordell combination allows. Naturally, many frequencies can create difference and odd-order products, satisfying our requirements for group 2 tests.
However, I was restrictive in my search for frequency combinations because I didn't want to give up the CCIF and Cordell test, nor did I want to make the instrument too complicated.
PHOTO 4: IM meter seen from the top. The peak-to-peak detector board is immediately behind the front panel, then the three filter boards, making up the analyzer part. Behind the analyzer are the four oscillator boards and the two transformers. The power supply board is to the right of the filter boards; the summing amplifier to the left.
I filled many sheets of paper with frequency combinations and calculations of even- and odd-order products, until I realized I should profit from the two frequencies already present: the 19kHz and 20kHz CCIF frequencies. After this, the puzzle solved itself: I had all three frequencies present for a three tone combination with a higher-than Cordell rate of change:
f1 = 9kHz, f3 = 19kHz, f4 = 20kHz, which will produce a difference frequency at:
f4 - f3 = 1kHz, and an odd-order product at:
f3 - 2f1 = 1kHz.
Obviously, these two tones would beat against each other and could cancel, so I had to offset one frequency to avoid this. Again, many combinations could be used. Keeping the 19kHz and 20kHz frequencies and offsetting the 9kHz to 8,950Hz would produce a difference frequency at 1kHz and an odd-order one at 1.1kHz; with 9,050Hz the difference frequency would again be at 1kHz, but the odd-order product would fall to 900Hz.
The necessary bandwidth of the analyzer would still be 200Hz, although the center frequency would have to be moved slightly.
The offset of the 9kHz moves the Cordell product frequencies as well, but they will coincide with the frequencies I just mentioned for my combination. With f1 = 8,950Hz, f2 = 10,050Hz and f4 =20kHz, the difference frequency is at 1.1kHz and the odd-order is at 1kHz.
With f1 = 9,050Hz, the two products are 1kHz and 900Hz, respectively, exactly the frequencies produced by my combination.
FIG. 17
The offset I finally adopted in my instrument is based on moving f3 (19kHz) 50Hz, and keeping f1 = 9kHz, f2 = 10.05kHz and f4 = 20kHz.
Again, both 19,050 and 18,950 will produce products at 950Hz and 1,050Hz, but with 18,950Hz, the even- and odd-order product frequencies will be the same with either the Cordell or my combination:
the even-order product at 1,050Hz and the odd-order at 950Hz. Since the product frequencies are the same, the only change required is the signal source, where the 19kHz oscillator must be tuned to 18,950Hz. Thus, you get an extra operation mode at no extra cost. You might wonder what this buys for you.
Well, the difference compared to the Cordell test is a higher rate of change:
1/3(9kHz + 18.95kHz + 20kHz)2x =0.100426V/ u-sec per volt peak
I also expect that, because of the higher frequency components, the composite waveform will reach this rate of change more often than does Cordell's.
Thus, the new combination will have a higher sensitivity in measuring high frequency related nonlinearities. How ever, it also has a drawback: according to G. Stanley et al, ? a three-tone expansion Continued on page 28
Continued from page 26 for cubic nonlinearity shows the odd order product of the type f3 -2fl has a coefficient of 3/4; and the triple-beat type, f4 - f2 - fl, has a coefficient of 3/2; so the new three-tone test is less sensitive to odd-order nonlinearities than the Cordell test. However, its ability to exercise high-frequency nonlinearities more effectively more than compensates for this in my opinion. It has proved itself especially sensitive to high frequency common mode distortion, as you will see in Part III's measurement section. More data will be needed, though, to be able to judge its validity as a general-purpose IM test.
Moving the 19kHz signal to 18.95kHz changes the standard CCIF test slightly and the difference frequency will appear at 1,050Hz instead of the normal 1kHz.
The decrease of the rate of change from 0.1225 to 0.12236 is about 0.1% and will not significantly alter the results of the measurements compared to the standard CCIF. However, if you are obliged to quote standard CCIF in your specifications, you are better off leaving the CCIF frequencies unchanged and offset the 9kHz tone instead, as described above.
Either way, the Cordell and the new combinations will yield the same results, but you must re-tune the band pass filter to the new center frequency when offsetting the 9kHz tone.
Measuring ‘Old ’
My measurement setup is shown in Fig. 17. Theoretically, the measurements can be carried out without an oscilloscope, but you should use one to monitor the IM products. I also monitor the composite signal from the amplifier to ensure no oscillations or overload are present, but to do this you need a dual-channel instrument.
The measurement procedure is:
1. Before you start any measurement, set the oscillator level to zero. Set analyzer input gain to zero decibels and peak-to-peak detector selector to 20V. Set analyzer ‘ CAL ’ pot to zero and Cal/Input switch to Input.
2. Monitoring the output of the amplifier, adjust the amplifier output to the desired peak-to-peak level with the oscillator level control. Switch to Cal position and adjust the peak-to-peak level to 2.82V DC with the analyzer in put pot.
3. Read the level of IM on the millivoltmeter. Sensitivity is 0.0001% per millivolt as read from the millivoltmeter. Thus you will have the following full-scale deflections:
IVES. = 0.1%
0.1VE.S. = 0.01%
0.01VE.S. = 0.001%
0.001VE.S. = 0.0001%
To read IM above about 0.7% (equivalent to 7V on the millivoltmeter and the saturation point for the filters in the analyzer), you must set your reference at a lower value. For example, if you set your reference at 0.282V p-p, 0.1VE.S. will be equivalent to 0.1%, 1V
-----------------------------
G ULTI-T caps and a high stability original frequency. If you distortion; it should read 2.
one the the R2 the just output offset to input and adjust the input to 0.5VI9k 50dB relative to the have a buffer at the trimpot, ft should be have a THD analyzer, less than 0.002% at Apply a sine wave to just tested) and check a scope to the output of with an output level be warmup adjust output unter to the output and are out of range for the adjust P0) ceramics. P1 the mic the urn voltmeter doesn't introduce noise into the analyzer. If it does, use your voltmeter between ground and This completes the chassis and apply a 1mV/lkHz sine put gain to 0dB and the CAL pot to until you read exactly 1VRMS at the pot P2 on filter F2 This sets the total connect the output oscillator output to 0dB, peak-t until you analyzer ding to less PM meter. Fu rtic$e.
S up THE I ER chassis. Necessary equipment for DC voltmeter and an oscilloscope. should use it to test the oscillators should be supplied from a ± 18V Test a PI boards before installing them in the testing: frequency counter, AC milli-voltmeter If you have access to a THD analyzer you and the summing amplifier. AN boards unregulated supply.
1. Oscillators. Connect the oscillator. With supply tween 5.5V and 7V RMS. level to 6.5V RMS with P2. adjust to the desired ment, pad the caps high-quality trimpot, board run for a few , P an AC miHivoftmeter and applied It should oscillate After a couple of minutes Connect the frequency co frequency with I . If you with small value COG (N highly stable both mechanically and electrically. Lat minutes and recheck the frequency; with COG cera within a couple of hertz of adjust P3 and C4 for minim all three frequencies. one of should be a
2. Summing amplifier. the inputs (you can use of the OsCi you the output for correct gain (see description In the article). With A1 and R2 in the circuit, the output equals iriputl2.75. (With 6.5V at the input the output should road 2.36V RMS.) With shorted, the output equals the input .82. If you have a THD analyzer check distortion at 20V p-p output. It should read the re of your THD analyzer.
Check all inputs for proper gain in both positions of the gain switch.
3. Filters Fl and F2. Test each filter separately. With the input shorted, ad zero volts with P1 . Apply a O5V/1 kHz sine wave to the gain of the fiRer with P2 to lOx (output 5V RMS). Change Hz and read the output. It should be down approximately O.5V input. Perform the same tests on F2. Since it doesn't input, it should be driven from a low l impedance oscillator Remember to adjust analyzer to 60dB.
(less than 6000 output impedance) for proper operation. the F2 gain after installation to set the total gain of the
4. Filter F3. Preset the combination of Al + P1 and R2 + P2 to 17.68k0 and Ri' + P1'andR2' + P2'to l4.46 kfl with an ohmmeter. Apply a 0.5V/lkHz Sne wave to the input; the output should read close to 5V. Change the fre .3dBofthe 1kHz should be down quency to 950Hz and 1 ,O5OHz; the output should be within 0 level. Change the frequency to 2kHz and 9kHz; the output recheck approximately 29dB and 59dB, respectively.
5. Peak-to-peak detector. Apply a IV RMS sine wave to the input and check the DC output with a ba DC voltmeter. It should read 2.82V p between the plus and minus output or 1.41V between ground and any of the outputs.
Carefully this after installation Into the analyzer and make sure your one of the output terminals.
setup procedure for the board8. Install the boards In the wave to the Input of the analyzer. Set In- maximum. Adjust gain output of the analyzer to 60dB. Remove the sine wave and signal source to the input of the analyzer. Turn I ga of the of the test maximum (approximately 20V p-p), analyzer input gain to detector attenuator to 20V p-p, and adjust the CAL pot p on the DC voltmeter. Read the output of the milli-voltmeter; it should be less than 1 .5mV, correspon 0.00015% distortion. This Is the residual distortion of the tests should be done on amplifiers, as described in the a read 2.82V on the AC than rther
------------------------------
....equals 1%, and so on, and you can measure up to approximately 7% on the 10V scale of the millivoltmeter. Beyond this there is no point in measuring IM.
4. Adjust the amplifier output to a new value, set the reference to 2.82 p-p again and read the IM.
5. If amplifier output exceeds 20V p p, set the peak-to-peak detector input level to 40V p-p, and proceed as before.
Don't forget to double the reading of the amplifier output because of the attenuator. Alternatively, if you monitor the amp output with an oscilloscope, you get the correct peak-to-peak reading from it.
Before I put my instrument to ‘work ’ in actual audio design, I made a number of ‘calibration ’ measurements to see whether I could replicate Mr. Cordell's results. The two test configurations I used are the unity gain inverter (gain =-1) and the noninverting unity gain amp or follower (gain = +1).
In the first tests I used the notoriously slow 741 op amp, to compare my results to Mr. Cordell's. The unity gain inverter test results are shown in Fig. 18a. There is good correlation between my measurements and Mr. Cordell's for 20kHz THD and the Cordell MIM (Ref. 3, Fig. 9). The difference in distortion values results from the higher slew rate of my 741 compared to his: 0.85V/ u-sec versus 0.65V/ u-sec. The additional two curves show the CCIF and the new three-tone test, both being less sensitive at lower levels, but both showing a steep rise in distortion before reaching the hard slew limit (13.8V p-p for CCIF and 17V p-p for the new test). The Cordell test doesn't detect the hard slewing condition below 20V p-p. The CCIF is sensitive only for even-order nonlinearity, so the 741 has asymmetrical distortion, especially when reaching the slew limit.
FIGURE 18a: Unity gain inverter test results.
The results of the unity gain follower test, again using the 741, are shown in Fig. 18b. Mr. Cordell didn't publish corresponding tests, but the results seem to be as expected: the CCIF and the new three-tone test show a good sensitivity for the considerable amount of common mode distortion in the 741. The slew rate being practically the same for my 741 in the inverting and the noninverting mode, the peak-to-peak voltage at which the curves show a steep rise appear to be the same in both modes of operation.
Now, I could (and did) go on trying ‘old ’ op amps to correlate my measurements with Mr. Cordell's and others', but I am sure you are more interested in looking at some modern devices. In Part Ill we shall continue in this manner. I have also dissected some of my own amplifier modules and will suggest a number of mods that I hope will result in even better sound. In the meantime I hope you build the multi-tone IM meter.
Good luck with your project.
FIGURE 18b: Unity gain follower test results.
ACKNOWLEDGMENTS
A special thanks to Mr. Robert R. Cordell for per mission to use his idea of multi-tone IM measurement for this construction project. Also I thank him for his help in sorting out my questions regarding the reference level used in his measurements and for evaluating some of my measurements.
REFERENCES
1. Cordell, Robert R., ‘A Fully In-Band Multi tone Test for Transient Intermodulation Distortion, ’ JAES, Vol. 29, No. 9, September 1981.
2. Stanley, G., and D. McLaughlin, ‘Transient Intermodulation Distortion and Measurement, ’ AES Convention Preprint 1308, 1977.
3. Cordell, Robert R., ‘Another View of TIM, ’ Parts I and II, Audio, February & March 1980.
4. Al-Nasser, Farouk, ‘Tables speed up design of low-pass active filters,' EDN, March 15, 1971.
5. Jung, Walt, Audio IC Op-Amp Applications, 3rd ed., Howard W. Sams & Co., 1987, p. 188, fig. 5.24.
6. Switched Capacitor Filter Handbook, National Semiconductor, April 1985, pp. 3-20 to 3-22.
7. Data sheets for LF356, LM148, LF13741, etc., Linear Databook 1, National Semiconductor.
Also see: