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

E-mail: Jack.Smith@cliftonlaboratories.com
 

 

To search within the Clifton Laboratories site, enter your search term below.
 

 


Home
Up
Documents
Book
Software Updates
Softrock Lite 6.2
Adventures in Electronics and Radio
Elecraft K2 and K3 Transceivers

 

 

July 2007 Updates

 

28 July 2007
  • Crispino, I5XWW, has supplied me with some Bourns LM-NP-1001-B1 600Ω : 600Ω transformers to evaluate for use with a Softrock.(Data sheet at http://www.bourns.com/PDFs/LMNPLP.pdf)  I've provided extensive measurements at my Softrock page, but my conclusion is that these transformers will work if you have the correct terminating impedance�not 600 ohms, but something closer to 10K. Their high sensitivity to terminating impedance, however, requires considerable experimentation with your particular sound card before providing 96 KHz bandwidth. And, depending on your sound card's input impedance, they may not be acceptable at all. Hence, I cannot recommend these transformers as "plug and play" for maximum bandwidth. For 24 KHz operation, almost any reasonable impedance is satisfactory, however.
  • I've mentioned on my lattice crystal filter page that anyone more than casually interested in filter design should have a copy of Handbook of Filter Synthesis by Anatol I. Zverev in their library, but lamented the fact that it was nearly a $300 book. Brad, AL4T, wrote to say that Zverev is now available in a lower cost paperback edition. Indeed, Amazon carries it for $58, which is only about twice what I paid for the hardbound edition in 1975. To repeat�this book belongs in your personal library if you are interested in filter design.  If you are not an engineer, don't be put off by all the equations. There are plenty of tables and discussions you can use to design filters without delving deeply into the mathematics. But, if you take the time and wade through the math, you will have an even better understanding of filter design. Thanks again to Brad for alerting me to the lower cost edition.
     
  • Graham, KE9H, also writes about my island pad capacitance measurements. I had assumed (always dangerous) that the thin PCB stock I acquired surplus had 1 oz. copper foil. Graham says it's almost certainly 0.5 oz foil  in its present form, as the additional copper for a 1 oz delivered finish is supplied during the electroplating process (traces and holes) as the raw stock is made into a completed PCB. Hence, instead of 1.4 mils copper thickness, it's 0.7 mils. (0.036 mm and 0.018 mm, respectively.) This changes my pad capacitance calculations slightly, as the fiberglass portion of the sandwich increases. I've revised the calculations accordingly on my Prototype page.

 

 

27 July 2007

Geoff, GM4ESD, has asked how much capacitance exists between an island pad and ground, assuming the pad is cut on a piece of double-sided PCB stock and that the top and bottom copper planes are tied together, as is normally the case.  The same question, of course, applies to conventional Manhattan-style construction with glued-on bits of PCB stock on top of a copper PCB surface.

I've added a section to my Prototype page presenting both theoretical and measured values for island pad capacitance.

 

26 July 2007

Because my shop has spread over most of the basement, it's common to move a particular piece of test gear from one end to the other. I've grown fond of scope carts (and own two), but several pieces of gear I own are too large for the carts I see on the surplus market.

There's a reasonably priced instrument cart available from Easy-Up (web site brochure at http://www.easyup-inc.com/rapidCat/WebPageImages/1271/3258/Easyup%20Mobile%20Cart.pdf) I now have five of these, two flat top and three slope top "instrument carts," acquired over the years. The dealer I've used is Electronix Express, and the carts are shown at http://www.elexp.com/am_easy.htm.  These carts are moderately reasonably priced, in the $71 to $85 range, depending on size and finish. I've found some on-line outlets selling these carts at $20 or more higher price, so shop carefully, if you decide to purchase one.

 The carts are available in either unfinished hard board or gray plastic laminate over particle board, with the laminate finish being a few dollars more expensive. All mine are plastic laminate finish. The carts have 3" casters for mobility. They come knocked down, and it takes about 20 to 30 minutes to assemble one. Normal hand tools are necessary.

Until recently, I've found the smaller size sloped cart adequate. However, the HP3562A Dynamic Signal Analyzer I recently acquired requires the large EZ45-8G cart, 23" wide x 19" deep. However, the 3562's feet are more than 19" spaced, so I added a 1/2" thick 23" x 24" secondary top to the EZ45-8G cart. I also added a blocking strip at the rear of the top to prevent the 3562A from sliding backward off the cart.  Normally, the Easy-Up carts retain instruments with a rear stop bar, made from steel tubing, with rubber pads. However, the 3562A is so large and heavy that the stop bar is not adequate, particularly after I relocated it to the rear of the cart, made necessary by the shelf extension.

The quality of these carts is not equal to a purpose-made Tektronix or HP oscilloscope cart, of course, but they are serviceable, particularly in the basement shop and the price is reasonable. The 3562A weighs 70 pounds or so and the cart is quite stable with it in place.
 

HP3562A Dynamic Signal Analyzer on the modified EZ45-8G cart.

I've added a secondary top to the cart, along with a wood stop strip at the rear.

I also had to relocate the normal stop bar.

 
EZ45-6G cart with an Advantest R3463 spectrum analyzer on top, and an Elecraft K2 on the bottom shelf, along with a Bird 100 watt 30 dB attenuator.

The EZ45-6G is 15" wide x 19" deep and fits most instruments.

You can see how the stop bar works with a typical size instrument. The bar rotates up and the rubber pad contacts the instrument's rear.

 
25 July 2007

Kees, K5BCQ, loaned me a Triad MIL-T-27E-SP68 audio transformer to measure, as a possible audio isolation transformer to use with a Softrock receiver. The transformer has, when properly terminated, a good frequency response, up to 100 KHz, but you must ensure it is  terminated with the 10K design impedance to avoid a nasty high frequency resonance. Some sound cards will present the required 10KΩ impedance to the transformer, whilst others, such as the E-MU 0202 unit I use, would require auxiliary terminating resistors, as the card's input impedance is 1 MΩ.

I've added the measured data, along with an LTSPICE model of the MIL-T-27E-SP68, to my Softrock page. Use the bookmarks at the top of the page to jump directly to the data.

 

23 July 2007

I've added details to the Softrock page on a broadband 1.8 MHz - 30 MHz 1:1 transformer I use to to isolate the antenna from my Softrock receiver.

The design may be of use for other purposes, of course.
 

22 July 2007

I've added a frequency response plot to my Softrock page of the Triad 5-58X 600Ω : 600Ω audio transformer that I use to isolate the receiver's I and Q channel outputs. Unfortunately, these transformers seem to no longer be available, but they have excellent performance.

 

20 July 2007

I received three diamond core drills this morning and tried them as PCB pad cutters. They work well, and better than my home made pad cutter. I'll add more discussion at the Prototyping page.
 

Pads cut with 5 mm (3/16"), 7 mm (1/4") and 8 mm (5/16") diamond core drills.

The amount of copper removed with the two 1/4" drills differs slightly with the depth of cut. It's important not to cut too deeply as there's a risk of removing all the copper if the disk is snagged by the drill.

 
18 July 2007

The last few days have been devoted to working with 21.4 MHz monolithic crystal filters from ECS, ECS-21K7.5A and ECS-21K7.5B, in particular. Click here for the data sheet. The 'A part is a two-pole filter in a single UM-1 case (similar to an HC-49/U crystal case, but smaller)and the 'B part is two matched two-pole filter modules, each in an UM-1 case. Both have a 7.5 KHz 3 dB bandwidth, although different skirt response, as one might imagine.

The data sheet would have you believe that these filters can be cascaded with a simple mesh coupling capacitor to ground between each pair of 2-pole section. Don't believe it. Or, at least I couldn't make it work and a telephone conversation with an applications engineer at ECS lead me to believe it was not possible either.  Rather, some isolation between filter sections is required. In the case of the 'B filters, each pair of 2-pole elements requires isolation. I believe isolation is also required between each 2-pole element if one is using 'A 2-pole sections as well.

Ideally, the isolation is provided by an amplifier with high IP3 and low gain. Only four or five dB is required to offset an 8-pole filter loss, which is about 2 to 2.5 dB per 4-pole group. For testing purposes, I used an 850 ohm, 6 dB resistive pad which provided enough isolation to do the trick.

I also tried L-networks, pi-networks and transformers to match the filter impedance with my 50 ohm test gear. Although all worked, the transformers worked the best. To match 50 ohms to 850 ohms for the 'B filters, for example, I used an FT37-61 core with 6 turns on the 50 ohm side and 25 turns on the 850 ohm side, for a theoretical 50 ohm : 868 ohm ratio. Plenty close enough.

Following the data sheet method, without isolation between filter sections, produced poor skirt rejection and a lumpy passband, as shown below. I could make the flank spur move closer or further from the center, but it would never move where it belonged, merged into the central passband.
 


With a resistive pad to isolate the two 4-pole filter sections, the response looks like it should�flat top and smooth flanks, as reflected below.

Note the price paid for the resistive pad. The insertion loss is now 11.5 dB, whilst in the earlier direct-connection design, the loss is about 4 dB. Hence, an isolation amplifier has a certain appeal as it would remove the excess loss whilst providing even better section-to-section isolation. This layout also has excessive filter blow-by which is probably related to the construction, most likely ground loop coupling.

The top sweep shows the flanks as being noise limited, while the lower sweep shows clear evidence of blow-by.
 

 
Prototype filter with resistive pad isolation. Construction is Manhattan-style.
15 July 2007

Perhaps the most common trait of ham operators is that of a pack rat. It's hard to throw away something that may be used some day. When searching in the garage today for a hose nozzle, I ran across a coffee can with a couple hundred 1/2 watt International Resistance Co., carbon composition resistors. These resistors were new around 1960, as I recall acquiring them not long after that date, and are unused. Hence, they should not have drifted due to conditions inside equipment.

I measured a selection of these resistors today and found that unlike wine, carbon composition resistors do not get better with age. More details at Carbon Composition Resistors.

14 July 2007

I've added a section to my Prototyping page discussing the "island cutting" alternative to Manhattan construction. It works well, and avoids the perennial problem of breaking the super glue bond while soldering to a Manhattan pad. The main disadvantage is that you have to plan the construction and cut all the islands at once, or at least that's the safest approach. Now, it may well be that this is not a disadvantage at all, but rather enforces proper planning. I'll leave that call to you.
 

21.4 MHz crystal filter board built with island cutting prototyping technique.
 
12 July 2007 1400 EDT

If you own HP test equipment, it may have a NiCd battery requiring replacement. Before nonvolatile EEPROM memory became available, HP (and others) used rechargeable NiCad batteries to back up instrument settings held in ordinary RAM. K8AQC loaned me his HP8116A digital function generator a couple weeks ago and I noticed it failed to recall the last parameter settings, and when powered on would default to 1 KHz sine. 

Opening the 8116A's top cover showed why it was not recalling the last parameters. This particular instrument was manufactured in 1986, and had the original 1986 NiCad battery, which had corroded and leaked from both ends. Fortunately, the corrosion was confined to the immediate area and had not damaged adjacent traces or ICs.
 

Both ends of the battery leaked and corroded the top solder pads.

Although I'm sure original equipment NiCad batteries are available, neither Mouser nor DigiKey stocked them.  Rather than search,  I made up an exact replacement from two Sanyo N-110AA NiCad cells obtained from Mouser. These are 110 mAh NiCad cells, identical in physical size to the two cells used in the original GE battery.
 
The original GE NiCad next to the two Sanyo replacement cells.

To make the replacement battery, I soldered the two Sanyo cells in series and encapsulated them with a short length of white heat-shrink tubing.
 
Replacement battery in place. The rectangular pads on the PCB's top surface have some residual corrosion damage. Fortunately, the bottom pads, where the tabs are soldered, were clean and no damage to the PCB was caused.
 
12 July 2007 0745 EDT

I've measured the Softrock Lite 6.2 10 MHz band receiver's input impedance and VSWR and posted the results at my Softrock Lite 6.2 page.

11 July 2007 PM

I've tried two software defined radio programs from I2PHD today, Winrad and SDRadio. Comments at my Softrock Lite 6.2 page.

I've also added bookmarks and jumps on the Softrock page for easier navigation.
 

11 July 2007

I've added a few paragraphs and photos to my prototyping page showing a BNC mounting bracket I designed and made today. The brackets were spurred when I started tinkering with RF breadboarding a "Mode H" switched mixer and found the small PCB holders I normally used too small for all the circuitry destined to be part of the experimental layout.

The "mode H" or switched mixer was developed by Colin Horrabin, G3SBI, and offers extremely high third-order intermodulation intercept figures. Readers interested in learning more about Mode H mixers might find Martein's (PA3AKE) page http://www.xs4all.nl/~martein/pa3ake/hmode/index.html a useful reference. As I have some actual results, I'll provide my measurements as well. I do not intend, however, to do the exhaustive optimization Martein has performed.

The bracket is for larger experimental layouts than are accommodated in the small holders I've made. It's inconvenient to get RF signals or audio signals in and out of a larger breadboard without making permanent cable attachments, so my thought was a small aluminum bracket to hold a single-hole mount BNC connector. The bracket can be attached to the edge of the PCB breadboard for a mechanically sound connection. I made 10 of these in about five hours today with the milling machine and lathe. It's not the highest quality machine work I've ever done, but these are more than adequate for the task.
 

BNC mounting bracket for Manhattan-style breadboarding on a scrap piece of board stock.

The intended use is for larger Manhattan-style layouts.


A variant design would mount the BNC vertically, so that it might be dropped into the middle of  Manhattan board, rather than at the edges as these brackets require.
 
10 July 2007

If your oscilloscope is to work to the limits of its specifications, it's important to use the proper probe. For most work, a 10X probe, with a frequency specification matched to your oscilloscope is suitable. Tektronix has an excellent tutorial on the care and feeding of oscilloscope probes at www.tek.com. Use the search tool at Tektronix's home page and search for "ABC of Probes." You may have to provide some registration information before obtaining access to the tutorial.

The probe tip itself is only half the equation, however, as the ground connection is also critical for the best results. I was looking at a 20 MHz logic level clock oscillator today and thought it would make a useful illustration of scope probe grounding. The oscilloscope is a Tektronix 2246, 100 MHz analog, with Tektronix P6106 10x probes.

 

Oscillator output with the normal long ground lead clipped on the oscillator ground pin.

Although the oscillator should have a flat top and bottom square wave with 5 V PP, we see a spiked waveform over 7.5 V PP.


The oscilloscope image above shows signs of major ground lead ringing or oscillation. The fast 20 MHz clock's fast rise/fall is shock exciting the resonant circuit consisting of the inductance in the 6" ground lead and stray capacitance. This is also known as "ground bounce."

We can reduce the ground lead inductance in the normal fashion, by shortening it. And, I don't mean using a 4" ground lead instead of a 6" lead. Instead, slip the hook sleeve off your scope probe. You should see a sharp pointed pin for the center lead with a surrounding metal shell exposed. The metal shell is the probe ground. (Some older probes do not have a removable sleeve. I have a couple of HP probes from the 1960's of one piece construction, but probes manufactured in the last 20 or 25 years are almost always of removable sleeve design, even inexpensive generic probes.)

Your probe kit may have come with a wire coil or spring type device, shown below slipped over the probe. This is a suitable short ground connection for medium speed circuits. (There are even shorter probe arrangements for faster transitions.)
 

Tektronix P6106 probe with sleeve removed and grounding clip in place.

If your probe set did not include a ground adapter of this type, you can easily make one from a short length of bare wire.
Home made spring ground adapter. I've soldered one end to the oscillator's ground pin. (It goes without saying, of course, don't solder to your scope probe!)
 
Clock output with the home made spring grounding adapter.

Although the on-screen readout says 5.87 V PP, counting graticule lines shows it to be closer to 5.0 V.

 


The improvement should be obvious. There's still a bit of ringing and tilt, but the waveform now looks like a square wave should.

If you are looking at an audio signal, or one at a couple MHz or less, a long ground lead won't usually be a problem. However, at higher frequencies and to see the best edges on a waveform with fast rise/fall times, you have to pay careful attention to your probe's ground connection.

One final reminder�even a 10X probe has a low input impedance at high frequencies, as it is shunted by a few pF capacitance. By the time you reach 30 or 40 MHz, even a few pF shunt capacitance represents a few hundred ohm impedance. The shunt capacitance isn't too important in looking at the 20 MHz TTL clock oscillator, but it can cause problems in other circumstances. Tektronix covers this issue, along with many others, in the "ABC of Probes." If you have not yet read it, do so.

 
09 July 2007

I've completed third order intermodulation measurements on the Softrock Lite 6.2 receiver, as well as minimum discernable signal and low frequency noise (around the 0 Hz IF). The data is at the Softrock page.
07 July 2007

I recently purchased a used General Radio GR 1658 Digibridge and have been experimenting with it today. The 1658 reads resistance, capacitance and inductance with quite decent accuracy, 0.1% in many cases. The test frequencies are 100 Hz and 1000 Hz with my machine (it is set for European frequencies) so it's not really suitable for very small capacitance and inductance values in the ranges we often use in RF work, but it's still quite handy to have around the shop.

In July 2001, I measured the effect of bias voltage on capacitance, some of which data is presented at my Capacitor Voltage Change page. Today, I've taken more data with the 1658 Digibridge and expanded the page to include more capacitor types (electrolytic and tantalum) and some smaller value (1000 pF) parts of the type we use in radio work, including  polystyrene, mica, C0G/P0 dielectrics. I've added the new data to the Capacitor Voltage Change page, now about double its prior length.

I've also tried KGKSDR software with my Softrock 6.2 with poor results�way too much CPU resources required to be usable. Details at the Softrock Lite 6.2 page.
 

06 July 2007 PM

It's been a busy day. The E-MU 0202 sound card arrived and I've had a couple hours to work with it and the Softrock lite 6.2 receiver. I've extensively modified (rewritten) the Softock Lite 6.2 page to reflect what I've found today, but the bottom line is that the E-MU 0202 provides better performance in several respects, including a maximum 192 kb/s sample speed. However, the added CPU resources required to process the higher sample rate, and the associated graphical activity, really tax my SX260's CPU resources, upwards of 85% CPU usage under certain combinations.

 

06 July 2007 AM

I thought I had damaged my 10 MHz Softrock lite v. 6.2 receiver by overloading the signal input with a +20 dBm signal during testing, so this morning I went through it with an oscilloscope, voltmeter and signal generator (at -10 dBm  this time). I found some odd signals and voltage readings centered around U4 (an FST3253 fast bus switch used as the mixer) and U5, a TLV2462 dual op-amp active filter. When building the board, I must have made a poor solder joint at U5, pin 3 or 5. Reheating both pins fixed the problem and the receiver seems to be working properly.

Later this morning, I should  receive the E-MU 0202 USB 2.0 external sound card to use with the two Softrock receiver boards.

Before  trouble shooting the receiver board, I made up a ground clip to header pin adapter. I acquired several of these along with some Tektronix scope probes, but I thought it would be a useful quick project to document on these pages.

The device I modeled it after is a "SAFTLEED" as shown below. The black projection is a receptacle for a standard 0.025" square header pin. The white object is an insulated wire connector that serves as a insulating shell to protect the wire loop. The purpose of this gadget is to provide a safe connection for an oscilloscope ground clip to a standard header pin. Most oscilloscope ground leads have an alligator clip that's too large to easily connect to a header pin without risking shorting adjacent pins.

To use the adapter, you slide the receptacle over the header pin and clip the probe's alligator ground clip onto the wire loop inside the shell. It works well, and can also be used for signal connections, as it's often more convenient than trying to clip the scope probe tip onto a header pin. I have one in red, one green and one dark blue, which makes distinguishing connections easier.

You can read the details in making one of the adapters, and see how it is used, by clicking here or going to the link Header Adapter at the top left of this page.

Original "SAFTLEED" as supplied with Tektronix probe set.
My home brew version of the SAFTLEED.
 
04 July 2007 0800 EDT

I realized last evening that the 21.4 MHz filter intermodulation data has a flaw--the two signal generators were spaced far enough apart that the intermodulation product fell outside the filter passband. Hence, an IMD product generated in the first filter section would be attenuated by the second filter section.

This morning, I re-ran the data with the two signal generators at 21.401 MHz and 21.399 MHz, i.e., 2 KHz spacing centered in the filter passband. The 3rd order products will thus be at 21.397 and 21.403 MHz, within the filter passband. We should, therefore, see IMD caused by any filter section.

I ran tests with filter inputs of +6 dBm, -4 dBm and -14 dBm, with the -4 and -14 dBm values obtained by inserting 10 and 20 dB attenuators between the combiner output and filter input.

To show the increase in intermodulation products, I ran  the spectrum analyzer in dual trace mode. The reference trace for each level is a 2 dB pad, which almost exactly duplicates the filter's 1.9 dB measured insertion loss. The second trace is the filter. Using a 2 dB pad allows the trace data to match and presents the spectrum analyzer with the same input level as when the filter is in place.

Spectrum analyzer data is captured with a Prologix GPIB-to-USB interface card, and rendered with KEFX's 7470 emulation program.

The blue trace is the saved 2 dB pad data, whilst the black trace is the 21.4 MHz filter. Where the two traces overlay each other, the blue will be on top.

The first test is with zero dB attenuation, which presents the filter with an input signal of about +8 dBm. Note that not only are 3rd order intermodulation products visible, so are 5th order products.

IP3 is: +23 dBm, reasonably close to the +26 of yesterday's data, but worse, as expected since the IMD product is no longer masked by filter attenuation. (IP3 based on filter output; add 2 dB if an input reference is desired.)

The IP3 of the test setup is +42 dBm, with the limiting factor being combiner isolation and mixing in the signal generators.
 

 

With 10 dB reduction (to -2 dBm) to the filter's input. Note that the 5th order products theoretically will drop 50 dB with 10 dB change in input level, which places them below the noise level in this particular test setup.

The filter's IP3 is +24 dBm.


The final plot is with 20 dB attenuation, so the filter input is -12 dBm. This is slightly below the filter manufacturer's -10 dBm recommended maximum input level.

The IP3 is +22 dBm.

IP3 conclusion.

Variations of ±1 dB can be disregarded for a quick test such as this.  Hence, the filter shows an essentially constant IP3 of about +23 dBm for input levels from -12 dBm to +8 dBm. The filter's IP3 values are well behaved and do not show significant changes in computed IP3 with different drive at the levels tested.

+23 dBm is not a bad IP3 level, but if to be used in a receiver with a target IP3 of, for example, +30 dBm, this filter would be the limiting factor.

 
03 July 2007 1800 EDT

In my 21.4 MHz crystal filter discussion earlier today, I mentioned that monolithic crystal filters are prone to intermodulaton distortion, particularly when driven hard, i.e., above -10 dBm. I've made a quick IP3 measurement on the crystal filter assembly this evening.

The equipment is:

  • Signal generator 1: HP 8640B
  • Signal generator 2: Panasonic VP8191A
  • 6 dB pads on both generator outputs Minicircuits CAT-6
  • Combiner: Minicircuits LRPS-2-1A
  • Spectrum analyzer: Advantest R3463

Both signal generators were set to +18 dBm, one slightly bellow the filter center and the other slightly above. The net loss through the 6 dB pads and the LRPS-2-1A combiner is a bit over 9 dB.

The spectrum analyzer image below shows the IMD floor of the test setup. The filter is replaced with a BNC barrel connector in this test.

The IP3 is approximately +43 dBm.


The figure below replaces the BNC barrel with the 21.4 MHz crystal filter. Note the significant increase in the two IMD spurious signals generated by the filter.

The IP3 is approximately +26 dBm, 17 dB worse than the measurement floor.
 

An IP3 of +26 dBm is probably acceptable for the application. For one thing, the the test drives the filter input at around +9 dBm, which is well above the level that will be applied in practice.  I'll run further  tests as I approach the final filter configuration, and see how non-linear the IMD products are with changes in drive level. A cubic law device has a 1:3 change ratio, i.e., 1 dB increase in input signal causes a 3 dB increase in 3rd order intermodulation products. I suspect the crystal filter will not follow a perfect cubic law and that with an input level more realistic, say -10 to -20 dBm, the IMD performance will be much better.
 
 
 
03 July 2007

I've taken a break from programming for the day, instead tinkering with a 21.4 MHz crystal filter. The next generation panadapter I'm working towards will use a 21.4 MHz 1st IF, with a relatively broad filter, 7.5 KHz or so, and a 2nd IF at 12 KHz with DSP. I'm uncertain whether to use a full I&Q 2nd IF or go with a conventional design and won't make that decision until further experimentation.

ECS has an attractive series of modular 21.4 MHz crystal filters. The filters are individual two-pole monolithic quartz units that may be cascaded in 2, 4, 6, 8 or 10 pole configurations. I built up a 4-pole 7.5 KHz bandwidth filter today,  and measured several key parameters.

These filters are intended for AM or FM communications (the 7.5 KHz bandwidth filters are likely intended for VHF-AM, such as air-to-ground communications) and as such have a flat top response. That's not what one wants for a spectrum analyzer, but the 21.4 MHz filter will serve as a roofing filter for the DSP, which will have a traditional Gaussian response. With wider DSP bandwidths, the response will be compromised, but still usable. I believe a 3 to 5 KHz Gaussian bandwidth will look good, but the maximum resolution bandwidth of, say, 7.5 KHz, will be more flat topped than rounded.

My concern with these filters has been ultimate rejection and flank bandwidth. With a 12 KHz second IF and without I&Q demodulation, image response can be a problem, particularly with wider bandwidths. I would like to see 80 to 90 dB at ±16 KHz from center.  That may be achievable, based on today's data, with a 6 or 8 pole filter.

A problem with monolithic filters is intermodulation, due to the crystal's parameters varying with drive, the so-called "drive level dependency." I will measure the filter's IMD tomorrow, although that may be challenging.

The three trimmers adjust the filter's input, output and inter-stage coupling. Tweaking these is necessary to obtain the desired passband response. The two toroids are part of an L-network to match the filter's 850 ohm impedance to 50 ohms.
 

Four-pole monolithic crystal filter. Each "can" has a two-pole filter. The second is mostly obscured by the shield but can be seen in part at the center of the shield.

The fixture is the one described at my prototyping page. I have a box of close to three dozen small circuits for this fixture, built over the last 10 years.

 

ECS-21K7.5 filter passband. Measured insertion loss 1.9 dB (spec is 2.5 dB with 1 dB ripple). I adjusted the trimmers for maximum flatness, which, I suspect, degraded the flank selectivity and slightly broadened the passband, at least partially shifting the filter from a Chebyshev to a Butterworth response. This is, by the way, the easiest crystal filter I've ever built, as ECS has done all the hard work.

The data is taken with an HP8752B vector network analyzer, time base locked to an HP 5065A rubidium frequency standard.
 

 

Group delay. Not too bad, about 75 μs variation in the passband.

 

 

Expanded span view. The flanks are symmetrical and show about 70 dB rejection at this scale. A wider span view shows ultimate rejection in the 90 dB range, which is the limit of the test configuration.
 
 

Over a 5 MHz span no signs of spurious responses. It will likely be useful to back up the 21.4 MHz crystal filter with an LC bandpass filter to avoid overtone response. To some extent, the input and output LC matching network provides low pass filtering, but additional LC shaping will be useful.
 
 

 
 

 

 
03 July 2007

Further research explains why Rocky consumes disproportionate CPU resources running on my Dell SX260. Although the SX260 has an Intel graphic controller on the mother board (the SX260 is a zero slot machine, so no substitute graphics card is possible), it does not have dedicated graphics memory. Rather, 32 MB of the main memory is allocated to the graphic controller. This cost savings measure extracts a major price on graphics speed. A graphic intensive program, such as Rocky, runs much slower than on a PC with a normal graphics card.

I've also ordered an external sound card with USB 2.0 interface to see how much of the noise and zero-Hz artifacts are related to the SX260's internal SoundMax sound card.
 

02 July 2007

Rocky v. 3.3.2 is now available. It fixes the S-meter problem. CPU resources are still an issue with my installation, and seem to be related to screen size. With the smallest reasonable screen, CPU requirements are 24%. That's high, but better than 53%.  With Rocky minimized to invisibility, CPU resources are 15%. I've added intermediate data points to my Softrock page. I've also run a quick test with an antenna isolation transformer and battery power for my Softrock receiver, without noticeable improvement.

More recent information ... Rocky loaded on my Gateway laptop is running around 10% or less CPU resources, even running full screen. My SX260 must be severely graphics-limited as the difference between the two computers is far more than might be attributable to CPU speeds or cores.
 

02 July 2007

I've installed  Rocky, v. 3.3.1, for Softrock receiver, and update the Softrock page to reflect my initial impressions of 3.3.1. Version 3.3.1 is a bug fix release, but the problems I observed are not addressed with v. 3.3.1.

 

01 July 2007 PM

I've received feedback on my Softrock page and have made changes to reflect those comments, and also my trial of PowerSDR software as well as Rocky. I like Rocky's user interface, but PowerSDR is not as demanding on CPU resources and also is (so far, at least) free of stuttering and crashing, unlike Rocky.

I'm now able to make an HP8116A function generator and an HP3456A voltmeter co-exist on the same GPIB bus. The data plot below of a 15 KHz DSP-based filter, is taken with that combination.

The responses at 5 KHz and 7.5 KHz are from the 8116A's third and second harmonics, respectively. Since the 3456A voltmeter is a broadband detector, it responds to any signals that make it through the 15 KHz filter. In this case, when set at 5 KHz, the 8116A's third harmonic is about -50 dBc, which is more than strong enough to be read. Even the second harmonic (8116A generator at 7.5 KHz) can be seen, and it's about -55 dBc. When measuring filter response with a broadband detector, you must keep generator harmonics in mind.

 
01 July 2007

The June 2007 Updates page is now at the June 2007 Archive and may be read by clicking here, or via the Archive links at the top of this page. It's now one year since I started the Clifton Laboratories web page.

The last few days have been busy with two activities.

First, one architecture I'm considering for the DSP-based panadapter is an I&Q down-converting design. Conventional swept spectrum analyzer front end, but the second IF selectivity and log conversion via DSP. This has some similarity to the Softrock receiver, so I purchased Softrock 6.2 Lite kits for 7 MHz and 10 MHz bands and built them over the last couple days. I've added a page with photos and some measurements, linked here or via the main navigation menu at the top left of this page.

I got a bit carried away making preliminary measurements on the 10 MHz receiver and blew the FST3253 bus switch mixer by over driving it with an HP 8640B signal generator. But, I've been listening to a 40 KHz swath of the 7 MHz band with the second kit since yesterday. I'm using VE3NEA's Rocky software and a Dell Opti-Plex SX260's built-in soundcard. It's not the quietest soundcard and the Softrock receiver board is laying on my programming table, but I can still hear a signal at 0.3 uV. Although the SX260 has 2 GB RAM and a 2.3 GHz processor, Rocky consumes about 50% of CPU clock cycles. Rocky is doing a lot of processing so it's understandable that it grabs quite a bit of CPU time. However, it makes other programming a bit sluggish at times.

The second project involves making a friend's HP8116A function generator play with a Prologix USB-to-GPIB interface card. Although the 8116A will accept message and properly set the instrument's parameters, reading back settings has been difficult, as the response strings are embedded in a sea of "NO MESSAGE" messages emitted by the 8116A. I've discussed the problem with Prologix's developer and a new firmware release (version 4.65) for the USB-GPIB adapter has solved the problem, or at least made it possible to easily read the 8116A's responses, while filtering out the NO MESSAGE fillers.

If you have test gear with GPIB communications, you should consider Prologix's adapter. It's reasonably priced and the times I've needed support, it has been provided very quickly and my questions/problems resolved in short order.