Clifton Laboratories 7236 Clifton Road  Clifton VA 20124 tel: (703) 830 0368 fax: (703) 830 0711

E-mail: Jack.Smith@cliftonlaboratories.com

Continuing in a series of measurements of my new K3 transceiver, this page examines the K3's receiver audio output.

Table of Contents
Introduction_and_Test_Setup_
High_Frequency_Noise_Characteristics_and_Artifacts
Line_Output_Level
Speaker_Output_and_THD
Headphone_Audio
SSB_Mode_with_Single_and_Dual_Tones
Headphone_Port_SSB_Mode_with_500_Hz_Tone_-_Harmonics
60_Hz_Hum_Pickup_versus_K3_Orientation
TTC-108_Transformer_Distortion_and_Intermodulation
TTC-108_Frequency_Response

Revision History
Originally written 30 August 2008
01 September 2008 - added single and dual tones within SSB passband
02 September 2008 - added LIN OUT = 2 and LIN OUT = 5 plots
02 September 2008 - added 60 Hz hum pickup versus K3 orientation
05 September 2008 - added section on TTC-108 transformer distortion and intermodulation
06 September 2008 - added frequency response data for TTC-108
06 September 2008 - modified to suggest 47R series resistance
06 September 2008 - modified to include single  tone at 500 Hz in 2.7 KHz bandwidth SSB mode
20 September 2008 - Added cross reference to non linear transformer page

Cross-reference: A considerably more detailed analysis of the Tamura TTC-108 transformer used in the K3's LINE OUT port is now available at my web page Non-Linear Transformer Behavior.

The figure below shows the test configuration for the K3 audio output tests. The only piece of equipment requiring additional description is the HP 3562A Dynamic Signal Analyzer. The 3562A is a dedicated Fast Fourier Transform instrument for audio spectrum and swept frequency response analysis over the range 0 to 100 KHz. (Actually, the lowest range is well under 1 Hz, close enough to 0 for our purposes.) I use it in preference to a computer audio sound card because it has  greater frequency range and has calibrated audio levels.

The GR1840A power meter has switchable load resistances. During headphone work, it is set to 50 ohms, the closest match to the 45 ohm headphones I normally use. Speaker measurements are made at 4 ohms and line out data is taken at 600 ohms.

For all  tests, the K3 was set  to CW mode, 600 Hz bandwidth, width centered at  600 Hz. The K3 is adjusted so that the beat note is centered using the K3's "CWT" function. AGC is enabled and set to fast.
 

My high frequency hearing is not good enough to know whether the measured artifacts I observed are audible to K3 owners. I invite comment on the subject.

In looking at other measurements described later on this page, I noticed that the K3's audio output has noticeably more high frequency noise and artifacts compared with other receivers, including Elecraft's K2  transceiver.

The 3562A spectrum analyzer plot below shows the K3's headphone (rear panel) output with the audio level set for a comfortable to loud audio level. The RF input is an unmodulated signal at -80 dBm and the K3 is in CW mode, 600 Hz bandwidth, width centered at  600 Hz. The K3 is adjusted so that the beat note is centered using the K3's "CWT" function.

Two points are of note. First, there are three discrete tone artifacts around 12 KHz. The center tone is at 12 KHz and the two stronger  tones are approximately 600 Hz above and below 12 KHz. Their level is about 45 dB below the CW  tone.  With a separation of 600 Hz, this strongly looks like signals related to the desired signal being aliased up and centered at 12 KHz feeding through. This is puzzling since the K3's 2nd IF, where the DSP work is done, is 15 KHz, not 12 KHz.

Second, note that the high frequency noise level is essentially constant from 2 KHz to 20 KHz, the limit of the band I looked at in this plot, as 20 KHz is normally the upper limit for human hearing.

Switching the signal generator's RF output off shows  the 12 KHz artifact, and also makes clear additional artifacts at just below 4 and 8 KHz. These are visible when the RF signal is applied as well, but are not as noticeable. The paired signals centered around 12 KHz are not present, leading further support to them being tied to the input signal.

I've carefully adjusted the K2's audio output level to match the K3's, within 0.5 dB or so. Both the K2 and K3 show some 3rd harmonic distortion of the 600 Hz beat note, down about 60 dB in both receivers, although the K3 is a few dB better than the K2 in this  regard.

More importantly, however, the 12 KHz artifacts are gone and the high frequency noise, above say 6 KHz is lower.

Elecraft's K3 Owner's Manual recommends operating with line output levels set below 10 as reflected in the excerpt to the right.

To assess the effect of various line out levels, I ran a series of spectrum plots with the HP3562A for LIN OUT settings ranging from 2 to 100 (2 plus 5 to 100 in steps of 5.)

Rather than present nearly two dozen plots individually, I've compiled them into an animated GIF image that steps through the LIN OUT levels.

The spectrum analyzer images are with an RF input signal of -80 dBm, corresponding to about S5. Stronger input signals will increase the audio levels and likely will increase harmonic distortion. There's also an interaction with the AGC slope and intercept as well, as those controls determine the demodulated audio level for any particular RF signal level.

Note that most of the distortion products are odd order, i.e., 3rd harmonic, 5th harmonic, etc. This is characteristic in several types of device distortion. Transformers, for example, will exhibit odd order distortion. Push pull audio amplifiers cancel even order distortion, so odd order products are more pronounced. An amplifier stage driven into severe clipping has an output resembling a square wave, which again is comprised of odd order harmonics. Of course, none of the cancellation effects are perfect.

Note that the artifact we observed in the headphone output around 4 KHz is quite clearly seen here as well. The frequency appears to be 3.9 KHz or so. It's about 60 db down from the desired signal, by the way.

The artifacts near the start of the spectrum analyzer plot are 60 and 180 Hz from the power line. (The K3 is powered for these tests by an HP E3610A DC power supply with very low hum and ripple. The power line signals are likely induced from nearby equipment.)

Since 10 is the recommended level, I'll provide a separate image for it, as well as  two lower settings, 2 and 5.

At LIN OUT = 2, the 3rd harmonic is down about 65 dB. (Note 15 dB/vertical division).

At LIN OUT = 5, the 3rd harmonic is down about 55 dB.

LIN OUT = 10, the third harmonic is stronger, down about 45 dB from the desired signal.

The K3's speaker output is  rated at 2.5 watts into either 4 or 8 ohms with a typical distortion of 2.5%.

Audio amplifiers generally produce maximum distortion into lower impedance loads, as the currents are higher for a given output power. Accordingly, I set the GR 1840A audo power meter for 4 ohms and took spectrum analyzer images for a selection of power levels. With -80 dBm RF level, the maximum audio output power I achieved was 1.25 watts.

The animated image below shows the spectrum changing with increased audio power. I had to change the spectrum analyzer axis during  this  test, so you'll notice an axis change for the highest power runs. The indicated power is in milliwatts.

The HP333A, like other analog distortion analyzers of its vintage, uses a narrow notch to remove the desired signal, with all remaining signal being considered to arise from harmonic distortion. At the 2.7% level, that's a reasonable assumption with the K3, but the K3's high frequency noise and non-harmonic artifacts will also be read as THD on the 333A's meter.

I also looked at the output distortion for varying headphone levels. The Sony headphones I use require only about 5 microwatts to provide a comfortable sound level. At 5 microwatts, the K3's distortion is quite low, with the first significant harmonic being about 70 dB below the desired signal.

The image below is of the headphone audio output at 1 mW, which is an earsplitting level with my Sony headphones. Less efficient headphones, of course, may require more output power. At  50 ohms, the maximum headphone output level I found with -80 dBm input was around 2 mW. Note how much the 3rd harmonic increases compared with the 5 microwatt setting.  It's still 50 dB down, which is more than decent, but the change is remarkable.

Recent comments on the Elecraft reflector caused me to look at the K3 (and at the K2 as a point of comparison) when operated with SSB filter bandwidth and with a single carrier and two carriers within the SSB passband. The diagram below illustrates the test setup.

Although my HP8657A signal generators are quite good performing equipment, both exhibit a group of 60 Hz powerline related spurious signals around their output frequency. I've discussed this at http://www.cliftonlaboratories.com/signal_generator_phase_noise_&_elecraft_k2.htm and reviewing this page first may help you understand some of the artifacts seen in this analysis. I've made some measurements in this section with an HP 8640B signal generator, which also has 60 Hz powerline related signals around its output frequency, albeit at a considerably lower level than either of the two 8657A generators I own. (The referenced web page shows some 8640B data as well.) In addition, the 8657A generators have some close in phase noise. With careful attention to the plots, however, none of these signal generator imperfections prevent accurate measurement of the audio performance of the K3 under the test circumstances.

The K3 is equipped with the stock 2.7 KHz bandpass filter and the selectivity was set at 2.7 KHz. I set the HP3562A for the range 0 - 2.5 KHz, except as otherwise noted. I also was careful to adjust the 3562A's input  gain to avoid overload that might contribute instrument artifacts.

The K3's LIN OUT gain is set at 10, the default setting, for these tests. The HP3562A has a 600 ohm termination across its input.

I've identified the main spectral features of the test plot below. The plot is of my K3 with the HP8640B signal generator as the sole input signal source, at -25 dBm, a very strong signal, S9 plus 40 dB or thereabouts, with the K3's preamplifier off. At this signal level, the K3's hardware AGC is operating.

• A are 60 Hz related hum signals apparently coupled into the test setup, although shielded cables are used and the 3562A is used in floating input mode to break  ground loops. These signals should be ignored in this and later plots.
• B are the  two spurious sidebands associated with my HP8640B signal generator. Note that these are about 68 dB below the carrier, which is not at all bad performance for a general purpose signal generator.
• C is the HP8640B's signal, as converted by the K3 into an audio beat tone output just over 500 Hz in frequency.
• D is the second harmonic of the beat note, down more than 60 dB from the fundamental (C).
• E is the third harmonic of the beat note, down 43 dB from the fundamental.
• F is the fourth harmonic of the beat note, down about 75 dB from the fundamental
• 1 and 2 are weak discrete signals of unknown source. They appear to be spaced equally from D, with approximately 60 Hz separation.
• The third harmonic at 43 dB below the fundamental is fully consistent with measurements I made earlier, reported on this page, on the LIN OUT port in CW mode. The headphone port performs much better if the level is kept low, demonstrating the third harmonic at -70 dB from the fundamental.

• The 60 Hz related hum products are considerably reduced, although the identical  test setup is in place. This suggests there is some coupling related to the K3.
• The second harmonic is not observable, indicating that it is at least 68 dB below the fundamental.
• The third harmonic can be seen at -67 dB from the fundamental, about 25 dB better than I see with the K3.
• Measurements reported earlier on this page demonstrate that the K3's LIN OUT output has higher harmonic production than seen in the headphone audio output.

The plot is unfortunately cluttered with spurious signals from the two HP8657A signal generators. I've identified the real signals.

• T1 and T2 are the two input signals. As the K3's dial is set, one signal provides a 500 Hz beat note and the second signal a beat note 370 Hz higher at 870 Hz.
• The third harmonic of T1's beat note (3x T1) is 50 dB down from T1 fundamental.
• The signal at the exact center (1250 Hz) marked IMD? could be a third order intermodulation product between T1 and T2.
• Additional discrete signals are visible at approximately 1875 Hz and 2250 Hz. These are not harmonically related to either T1 or T2 and might be higher order intermodulation products. I've identified them as IMD? but their origin is unclear. I don't believe  they are signal generator spurious signals, as the 8657A spurious signals drop in level as they space out from the center frequency. The signals shown with the IMD? mark are relatively constant level.

I've received a request to look at the K3's headphone port audio in SSB mode with 2.7 KHz bandwidth and a single tone at around 500 Hz.

The test conditions are: USB mode, BW 2.7 KHz FC 1.5 KHz, AGC-S, signal level into headphones set to a comfortable listening level with Heil Pro Set. RF level into K3 -40 dBm. The audio level is -29.4 dBVrms across a 200 ohm resistor used as a substitute for the Heil headphones, or about 34 mV.

The 3rd harmonic is down 76 dB, which is quite impressive. I'm not sure if the little blip just below 1 KHz is the 2nd harmonic or just a noise fluctuation. If it's the 2nd harmonic, it's down close to 80 dB from the fundamental.

I put the X marker (the round dot) on the 3.9 KHz artifact. It's about 78 dB down from the tone.
 

Jim, K9YC, suggested in a post on the Elecraft reflector, that the TTC-108 isolation transformers used in the K3's LIN OUT port, were unshielded and thus subject to hum induced via magnetic coupling. The transformers have a directional hum pickup characteristic,  by the way.

I found that this is indeed the case with my K3, and that by carefully rotating it on my workbench the hum levels would change 15 dB consistently and, with just the right orientation by almost 20 dB, although that was not stable.

The image below shows my K3 in the normal position, front panel parallel with the front of my workbench. (The only connections to the K3 are DC power and the HP3562A dynamic signal analyzer.) This turns out to be almost the worst case for hum pickup in my shop.

Rotating my K3, I can knock the 60 Hz noise down 15 dB, although the 180 Hz noise only drops about 6 dB. The only difference between the plot above and the plot below is the K3's orientation.

Cross-reference: A considerably more detailed analysis of the Tamura TTC-108 transformer used in the K3's LINE OUT port is now available at my web page Non-Linear Transformer Behavior

The K3's audio amplification section is capable, according to the active device specification sheets, of excellent performance. My measurements indicate that the K3's LIN OUT signal is not living up to the expected level of performance, and not by a small margin.

The K3's LIN OUT port is driven from a separate DAC,  U29 in the K3's schematic, a Texas Instruments TLV320DAC23,  with quite good specifications. (Elecraft uses two independent DACs; one for LIN OUT and a second for headphone and speaker output.) THD is -60 dB (0.1%) at levels of 10 mW.

This specification is for a 32 ohm load. In the K3's current configuration, U29's headphone amplifier sees at least a 600 ohm load at all times. Hence the actual output power is considerably less. With 3.3 volt supply voltage, and a small allowance for saturation at both power and ground rails, we might expect a 3.0 V PP output from U29's headphone amplifier. Into a 600 ohm load, 3.0 V PP corresponds to 1.8 mW. Of course, the limiting factor is going to be clipping at the positive and negative excursions, not power dissipation.

U29's integrated headphone amplifier output has separate sections for left and right channels and appear at the K3's rear panel through the following arrangement. (Right channel is shown; left channel is identical.)  C15 is a blocking capacitor as U29's operates with a single ended 3.3V power supply and hence its output is biased at V/2 or 1.65 volts. R20 is a 604 ohm resistor with a purpose that is unclear to me, but which turns out to be the source of significant performance impairments. The common side of the left and right transformers are tied together and connected, through RFI suppression chokes and bypass capacitors L3 and CN1. The high side of each channel is routed through a similar RFC / bypass arrangement.

Of these specifications, the most important is number 11, the "total harmonic distortion," quoted at 0.5%. 0.5% THD corresponds to harmonic suppression of 46 dB. Strictly speaking, of course, it's "total" harmonic distortion, i.e., the sum of all harmonics. However, usually one harmonic (the third) is much stronger than others, so that it dominates the 46 dB suppression specification.

To obtain a sense of how the K3's LIN OUT audio chain performs, I purchased a couple of TTC-108 transformers and built a simple test driver circuit. Duplicating the K3's DAC isn't necessary; U29's headphone output port is, for our purposes, a low impedance output audio amplifier. We'll use a Microcircuits MCP6021 op-amp as a substitute. The test circuit I built is illustrated below. It's a standard unity gain buffer amplifier with two inputs. Deserving additional explanation is the  bias circuit. In order to run from a single supply voltage, the non-inverting input must be biased at Vdd/2 and also the inverting input requires Vdd/2  reference voltage as well. Obtaining Vdd/2 can be done with a resistive voltage divider, but the MCP6021 (in the 8-pin DIP package, at least) includes a voltage divider with Vdd/2 available at pin 5. Hence, all that's necessary  to obtain Vdd/2 reference bias is to add C1, a bypass capacitor and connect Pins 3 and 5.

Additionally, since the inverting input also should be a Vdd/2 in the event of no input signal, R4 provides the necessary reference. Pin 2 is, for AC purposes, a virtual ground, so inputs 1 and 2 are isolated from each other, permitting us to connect two signal generators without worrying too much about interaction between the two.

It's important to keep in mind that the output of an op-amp with feedback such as this mixer circuit, is essentially zero ohms. The op-amp acts as a perfect voltage source, as the feedback keeps the output voltage constant. Of course, the op-amp can only provide so much voltage or power, but within these limits, and at the frequency range we are using it at, it has a zero output impedance. The headphone amplifier within the TLV320DAC23 has the same zero output impedance characteristics. Hence the simple MCP6021 mixer and buffer amplifier circuit below serves as a good analog of the TLV320DAC23's headphone amplifier.

To demonstrate the difference, and to show that the K3 is capable of with a simple component value change in the LIN OUT circuit, some data is  taken with R6 in  the  test circuit in place (duplicating the K3 configuration) and other data is taken with R6 bypassed, demonstrating what a revised K3 is capable of.

The data is taken using two audio signal generators, an HP 200CD and an HP3312 function generator. Both have relatively good harmonic performance, with the 200CD's harmonics typically 60 to 65 dB below the fundamental. The 3312 is not quite as good. Hence, when looking at the test data, in many cases the limiting factor in assessing performance is the harmonic content of the signal generators, not the MCP6021 circuit.

The data is analyzed with an HP3562A dynamic signal analyzer in spectrum analyzer mode.

The result is a sea of intermodulation products out of the  transformer (green trace). The blue trace, directly out of the op amp shows the signal generator harmonics, around -50 dBc as my signal generators take a bit of a performance hit in this frequency range as well. Note that the output level is relatively low here— -30 dBV, which is 30 mV or so per  tone (RMS). No intermodulation products are shown in the op-amp output; all those in the green image are generated in the transformer.

The output tone level is -22 dBVrms, or just under 100 mV RMS into a 10K load. Note that the 3rd harmonic is -45.6 dB with respect to the fundamental. That's almost identical with the 3rd harmonic suppression I found in the K3 once the LIN OUT level gets up in this range. The level of harmonic suppression depends on the fundamental signal level, of course, but the range I observed is on the order of -50 to -45 dBc.

• The source of the interference is the TTC-108  transformer.
• Operating the TTC-108 with the 604 ohm series resistor as in the K3's design is the reason the TTC-108 demonstrates the level of harmonic and intermodulation products seen.
• Replacing the 604 ohm series resistor in the K3 with a 47 ohm parts should significantly improve the LIN OUT audio performance. [Note: I had thought a 0 ohm part would be suitable but I've received a report that 47 ohms will provide better results.]

Note that I have not tried the modification directly on my K3, but the op-amp results give me high confidence that an improvement will result from this change. Whether the full degree of improvement seen in my tests will be seen in the K3 is unknown.

I also note that the TTC-108 transformer has a total winding resistance of 100 ohms (44 ohms in one winding and 56 ohms in the other winding) so if the 604 ohm resistors are replaced by 47R parts, no additional safety resistors should be required at the TTC-108 output.
 

I also measured the frequency response of the TTC-108 transformer when driven in the K3 arrangement with 620 ohms in series with the primary and when driven directly by a zero ohm op-amp. Before looking at the frequency response plots, let us first consider why a transformer has both an upper and lower frequency limit.

I've also discussed transformer frequency response in detail at http://www.cliftonlaboratories.com/softrock_lite_6_2.htm.

The schematic below is a transformer model showing the components that cause transformers to be less than perfect devices. (This is a simplified model but quite adequate for our purposes.)

It is possible to model a transformer and consider the imperfections as separately acting in the primary and secondary. However, the model may be greatly simplified if all of the imperfections are considered to be in the primary, as in the model depicted above. For example, suppose the secondary has 10 ohms winding resistance and the primary has 20 ohms winding resistance, and that the  transformer in this example has 50 primary turns and 100 secondary turns. The secondary's 10 ohms winding resistance appears to the primary as (100/50)^2 * 10 = 40 ohms. Rw, in this example, is 20 ohms direct primary resistance plus 40 ohms secondary resistance reflected back to the primary, for a total of 60 ohms. Likewise, the secondary's leakage inductance, etc. can be reflected back to the primary and combined with the primary's parasitic components.

The elements of the model are:

• Zs - the impedance of the signal source
• Lleakage - leakage inductance, i.e., some of the lines of flux generated by the transformer primary escape, or "leak," and do not cut the secondary windings. This leakage flux can be converted to an equivalent series inductance, Lleakage.
• Rw - the transformer's winding resistance. (AC resistance.)
• Cd - distributed capacitance across the primary.
• Rp - shunt resistance across the primary, due to dielectric and core loss.
• Lp - the magnetization inductance, i.e., the inductance due to the flux that links the primary and secondary windings.
• Cs - capacitance between the primary and secondary windings.
• Zl - load impedance.
• An ideal transformer with turns ratio N.

As said earlier, these parasitic components are shown on the primary side only and represent the net contribution of both primary and secondary parasitic elements.

Looking at the model should reveal why the frequency response rolls off below some frequency limit. Zs forms a voltage divider with Rw and Rp paralleled with Lp. As the frequency drops, the inductive reactance Xp of Lp drops and hence the voltage available to the secondary drops.

Likewise, it should be obvious why a low Z source will improve the low frequency response. If Zs is zero ohms, the voltage divider effect has reduced, with only Rw and Rp || Lp having an effect.

The data is taken with 10K secondary load, so the secondary winding resistance does not enter into the current calculation.

Good question. The frequency response data I've taken is into a relatively high impedance load, 10K ohm, representing a  typical computer sound card. (The HP3562A Dynamic Signal Analyzer I used for these tests has an input impedance of 1 Mohm, so I shunt it with a 10K ohm resistor during these sweeps.) Operating into a relatively high impedance load negates much of the effect of the leakage inductance and winding capacitance effect upon high frequency response.

At the high frequencies, the Lp drops out of the picture and the same voltage divider concept picks up new elements, the transformer's leakage inductance and the distributed winding capacitance. These parasitic elements are, in effect, in series with the load resistance and hence effectively limit the high frequency response only when their reactance becomes an appreciable fraction of the load resistance.-