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

E-mail: Jack.Smith@cliftonlaboratories.com

Table of Contents

IF_Noise_Blanker
Condition_1._Test_signal_present,_interfering_pulse_generator_disabled.
Condition_2._Test_signal_present,_interfering_pulse_generator_enabled.
Condition_3:_Test_signal_present,_interfering_pulse_generator_enabled,_and_IF_NB_on_(setting_Med-4)
Filter_Bandwidth_
dBV_Function_Linearity
2.7_KHz_SSB_Filter
500_Hz_5-pole_CW/Data_Filter_
FIR_versus_IIR_Filters
Time_Domain_Response;_DSP_Delay_and_IIR_versus_FIR_Filter_Response
DSP_Delay_Measurements
IIR_versus_FIR_
Effect_of_Added_DSP_Functions

Prepared 22 August 2008
Rev. 23 August 2008 - Added 600 Hz DSP / 500 Hz Crystal filter to frequency response plots

I'll start with the IF noise blanker because it's the shortest discussion.

The K3's Owner's Manual describes the noise blanking features built into the K3:

The output of the two RF generators are combined with a hybrid combiner and fed into the K3's antenna port. Audio from the K3's Line Out port is connected to either a Tektronix TDS430A digital oscilloscope or the built-in sound card of a Dell M6300 laptop computer. Trace images from the TDS430A are captured via a GPIB connection, using a Prologix GPIB-USB adapter and KE5FX's excellent 7470A plotter emulation program. I highly recommend both the Prologix 4.2 adapter and the 7470 emulation software. I use an old version (5.2) of Spectrogram to record the sound samples, and version 16 for spectrum analysis views. Version 16 does not support recording rates below 22 kb/s, so it's more efficient to use the earlier software to capture bandwidth limited data such as that out of the K3 under these conditions.

The K3 is tuned to 10000 KHz, USB mode, 2.8 KHz bandwidth, pre-amplifier off, AGC slow mode.

In case you're wondering why I selected 86.2 Hz as the PRF, I simply turned the PRF adjustment until the noise sounded the worst and measured the result.
 

The images below show the audio waveform out of the K3, followed by a spectrum plot of the output signal. To listen to 10 seconds of the audio recorded under these three conditions, click on either the audo waveform or spectrum analysis plots. (Note: the horizontal sweep rate is faster in the first oscilloscope plot than in the next two.)

Condition 1. Test signal present, interfering pulse generator disabled.

Note that the pulse peaks charge the slow AGC system and suppress the audio during the interval between pulses. (The K3 has an optional AGC mode setting to reduce this effect, but in the interest of a direct demonstration of the noise blanker, these tests are run with normal slow AGC.)

Condition 3: Test signal present, interfering pulse generator enabled, and IF NB on (setting Med-4)

(The audio frequency looks higher because this and the above oscilloscope images were captured at 5 ms/div, whilst the first image is at 2 ms/div.)

The K3 has two bandpass filter elements in series:

• At the first IF, 8215 KHz, are crystal filters. These are known as "roofing filter" as they are the first narrow filters in the signal chain and thus provide a "roof" over the subsequent stages to protect them from strong unwanted signals. The K3 supports up to five roofing filters.
• At the second IF, 15 KHz, the K3 implements bandpass filters digitally in the digital signal processing section. Thus, in order to make it to the audio output, an unwanted signal has to pass through the rejection supplied by both the 8215 KHz crystal filter and the 15 KHz DSP filter.

The K3's control firmware automatically selects the appropriate roofing filter based upon your choice of DSP filter. (It's possible to override the selection, of course.)

The main receiver in my K3 has three roofing filters:
   • 6 KHz for AM reception/transmission
   • 2.7 KHz for SSB
   • 500 Hz (5 pole) for CW and Data

If, for example, I'm in CW mode and have the bandwidth set at 800 Hz, the K3 will switch in the 2.7 KHz crystal filter and set the DSP filter for 800 Hz. However, if I reduce the bandwidth setting to, say, 400 Hz, the K3 switches to the 500 Hz crystal filter and sets the DSP to 400 Hz.

The traditional way one measures filter bandwidth is to turn the receiver's AGC off, connect an RF signal generator to the antenna input port and center the receiver on the RF generator signal. Adjust the generator level so that the receiver audio output is not clipping and the RF stages are not saturated. With an audio voltmeter connected to the receiver's audio out, note the reading. Adjust the signal generator up or down a known amount in frequency and increase or decrease the signal generator level so as to maintain the same audio  reading as when centered. Note the signal generator RF level and subtract from the center reading to obtain the filter attenuation in dB. Plot.

This process can be automated with a signal generator capable of computer control. I've used that approach in  examining the K3's AGC performance, for example. http://www.cliftonlaboratories.com/elecraft_k3_agc_and_s-meter.htm.

dBV Function Linearity

However, the K3 includes a very useful utility that measures the audio level with a 0.1 dB resolution. The reading is shown in the "B" display and can be read by an appropriate command over the K3's serial interface. This suggests an approach to let the K3 self-characterize its filters. Apply a clean, low phase noise test signal to the K3's antenna port. With the AGC off, set the test signal level so that the K3 is close to, but not exceeding,  the point of non-linear response. Then step the K3 over a range of frequencies by computer control, reading the dBV response for each step. Save the results to a data file and plot.

This setup is shown below.

Using the same test setup described at http://www.cliftonlaboratories.com/elecraft_k3_agc_and_s-meter.htm for automated AGC tests, I looked at my K3's DBV response. There's a  good 60 dB of linear response available, from -120 dBm to -60 dBm with the pre-amp on. With the pre-amp off, the linear range is -110 to -50 dBm.

HP specifies the 8657A's amplitude as better than ± 0.5 dB over the range -127 dBm to +7 dBm, so it's a tossup as to whether the ± 0.1 dB differences we see in the plot below are to be ascribed to the K3's DBV function, or error in the 8657A's attenuator, or, more likely, a bit of  both.

As far as absolute accuracy goes, it's immaterial for this purpose. We are concerned with signal levels relative to the reference point, i.e., the signal level when at the peak filter response.

In order to view the response of just the crystal filters, it's necessary to increase the DSP bandwidth well beyond the crystal filter bandwidth. In the case of SSB filters, the crystal filter has a nominal 2.7 KHz, and the maximum total bandwidth setting I could make work is 4.0 KHz. (There's probably a way to increase the SSB bandwidth setting beyond 4.0 KHz, but I'm still working my way through all the K3's complexities.)

The plot below shows the response in USB and LSB mode. There's only one SSB crystal filter, and the USB/LSB selection is done in the DSP stage, of course.

Since there's only one crystal filter for both USB and LSB, we expect the filter responses to be mirror imaged, and it appears they are within the limits our measurement process.

The anomaly at 10000 KHz is probably related to the K3 switching RF input bandpass filters at 10 MHz.

I measure the 3 dB bandwidth at 2.61 KHz, close enough to the 2.7 KHz nominal value for our purposes.

Both filters show a bend in the response along the outer flanks that seems to be due to the DSP filtering action. In order to see only the crystal filter response, it would be necessary to set the DSP filter width to 6, or preferably 8 KHz, bandwidth.

The plot noise floor is about 68 dB down from the filter peak. This does not represent the ultimate filter rejection, but rather the dynamic range of  the dBV functionality. Broadband noise is the limiting factor here.

With the 500 Hz crystal filter, we can see both the crystal filter and the DSP filter working together. I've provided plots showing both the full range and just the filter nose.

The blue trace is with the DSP set at 4000 Hz, and the 500 Hz crystal filter selected. (Normally the K3 would use the 2.7 KHz crystal filter when the user sets 4000 Hz bandwidth in CW or data modes, so it's necessary to disable the 2.7 KHz filter via software option when making this measurement.)

We see the crystal filter has a bit of bandpass tilt. I've designed and built a few crystal filters as well as LC filters and bandpass tilt is not uncommon. In this case, it's perhaps 0.7 dB or so, which is not significant for our purposes.

The 500 Hz, 5-pole crystal filter's key measured parameters are:

Note that with the DSP in play (500 Hz and below) the filter flanks become almost vertical, with a very significant drop in signal level for just a few Hz change in frequency, a very useful thing indeed. [The 600 Hz data is from a later run with a different DBV reference, so I  manually offset the peak level to coincide with the earlier runs. This results in the out-of-band levels being 5 dB or so higher than expected.]
 

For the two narrowest filter positions, 100 and 50 Hz, the K3 provides the user with a choice of IIR or FIR filter characteristics. The K3's default is that all DSP filters use an IIR implementation, but a menu item permits the 100 and 50 Hz filters to instead use FIR implementation.

 IIR, or "Infinite impulse response", and FIR, or "finite impulse response" filters differ in how they respond to narrow impulse signals. As their names suggest, an FIR filter has a finite response, i.e., after the inpulse signal ends, at a well defined, finite time later, the filter output has decayed to zero. An IIR filter, in contrast, theoretically has a never ending, infinite length response to a narrow impulse signal. IIR filters can provide steeper rolloff with frequency than an FIR filter for similar computational complexity.

As a practical matter, this does not mean that a noise spike into the K3's 100 Hz filter in IIR mode will be be audible a week from next Tuesday. Rather, we are talking about milliseconds or tens or hundreds of milliseconds in practice. The next section of this page, Time_Domain_Response;_DSP_Delay_and_IIR_versus_FIR_Filter_Response will explore how the two filter choices respond in the time domain to short RF bursts.

If you wish to learn more about DSP, IIR and FIR, Dr. Steven Smith (no relation) has written an excellent book, The Scientist and Engineer's Guide to Digital Signaling Processing. It's available in either a free web download version at http://www.dspguide.com/ or as a printed book. The printed  book is updated and expanded and is well worth the modest—for a technical book, that is— $48.48 price (based on Amazon.com pricing on 22 Aug 2008.) It's as non-technical as one can get with a book on DSP that's still useful. I own a copy of the printed book and refer to it frequently.

The plot below shows the frequency response of the K3's 100 and 50 Hz filters in IIR and FIR implementations. The data is taken with 2 Hz step increments and might benefit from 1 Hz steps.

The K3's Owner's Manual says:

Narrow DSP Filter Types

For bandwidth settings of 100 Hz or lower, the K3’s DSP normally uses a type of filter that minimizes ringing: the Finite Impulse Response or FIR filter. If you’d like steeper filter skirts, and don’t mind a small amount of ringing, you can select Infinite Impulse Response” or IIR filters for these bandwidths. Locate CONFIG:FLx BW menu entry, then tap 7 until you see IIR ON. Both main and sub receivers will use the same setting.

As the plot below demonstrates, there are some anomalies here. At 50 Hz, for example, the FIR filter skirts are steeper than the IIR filter. In the 100 Hz case, there isn't as much difference but the FIR filter skirts are a bit steeper than in the IIR case. What makes the filter skirt comparison difficult is that the IIR and FIR filters have quite different 3 dB bandwidths, at least for 100 Hz filters. Perhaps in this case, a table is worth 1,000 words, not a picture.

 In both cases, the FIR filters have better 60:6 dB shape factors, so I'm puzzled by the statement in the manual. In collecting the data, I thought that perhaps I inadvertently reversed the IIR and FIR options, so I took a second set of data which is essentially identical with the first set.

So far, our filter analysis has been in the "frequency domain," i.e., we've looked at how the filter behaves when excited by a stead state sine wave that is stepped in frequency.

However, the filter's "time domain" response is also of importance—how does the filter respond to a time varying signal, such as a fast rise CW element or a data signal. Intertwined with the time domain response is the DSP's delay. How long does it take between an RF signal input and audio output from the DSP?

There's a closely related parameter called "group delay" that is a measure of how the delay changes as the frequency of  the applied time varying signal changes. I have not measured the group delay of the K3's filters as it is a non-simple task.

The figure below illustrates the test setup for measuring DSP delay and filter step response. The SG-100 function generator outputs a fast rise RF pulse, 20 ms duration, with 100 ms repetition rate. A synchronization pulse from the SG-100 triggers the TDS-430A digital oscilloscope. Hence the time differential between the leading edge of the synchronization pulse and  the K3's audio output is the DSP delay. More accurately, this interval represents the K3's end-to-end delay including time delay in the analog stages. However, experience with analog and DSP-based receivers suggests that the nearly all time delay is attributable to the K3's DSP stage.

A 20 ms RF burst corresponds to the dot rate at 60 WPM.

We can see a small overshoot on the leading edge of the waveform (CW mode, AGC is on, slow mode, during these tests) due to the finite AGC attack time.

If we switch to IIR response, the 50 Hz filter has a rather distinct tail. Note that I had to change the horizontal sweep to 10 ms/div to fit all the tail into the image. The 20 ms main burst now comprises a 40 ms or so main tone burst plus a 20 ms reduced amplitude tail. One might argue that the tail is more like 40 ms, but the amplitude diminishes considerably after 20 ms.

As we engage additional DSP functions, the overall delay increases. The baseline delay is 17 ms, with no special functions engaged.