My first network analyzer was designed around 1985. It used a general purpose analog VCO (an Intersil ICL8038) as the driving source and a pair of IC RMS-DC converter chips as receivers. I can remember the date because I programmed it through some A/D and D/A converters from my Apple II computer. It had it's limitations, the dynamic range was only about 30 dB and the frequency topped out at 100 kHz. Good enough for audio, but not even close for the simplest 455 kHz IF.

This design started out a few years ago when Numerically Controlled Oscillators (NCO's also are called Direct Digital Synthesis or DDS chips) started appearing in designs [1] based on the Harris Semiconductor HSP45102 DDS Chip. This part allowed a two chip solution to generating sine waves from 0.01 Hz to around 16 MHz with 32 bits of digital programming. Wow, to build a low noise PLL to do this would involve lot's of bench debugging time, and be slow to change frequencies because of the low frequencies involved.

In 1994 I built my first of several DDS oscillators based on the Harris chip. When they are clocked at 40 MHz, 10 Hz to 10 MHz programming is very clean and reasonably spur free. The output can be coaxed to 16 MHz output, but Nyquist sampling limitations and aliased spurs start to be only -20 dBc down from the main carrier. For more information on DDS methods and limitations see the references at the end of this article.

Late in 1996, while working on a 60 kHz, WWVB receiver I was experimenting with a Philips, NE604A IC. This part is essentially a 80 dB LOG FM IF strip. I was using it because a feature of the chip is it's Received Signal Strength Indicator or RSSI output. The RSSI output gives a fairly accurate voltage that is proportional to the input IF voltage over a -20 to -100 dB range, that's 80 dB or so of dynamic range from a single low cost IC. While experimenting with the part I found that while it was designed for narrow bandwidth 455 kHz or 10.7 MHz IF applications it would work with a dynamic range of 60 dB or so in a wide band mode.

The connection was made, using a Harris based DDS chip and the NE604A I could now build a network analyzer that had 50 dB of dynamic range and a frequency range up to 16 MHz. This would be suitable for nearly all of my IF work and plenty of dynamic range would allow accurate measurements of high loss filters and high gain amplifiers. So this design was born.

Basics Of A Network Analyzer

A basic network analyzer is designed to show graphically, a plot of the voltage gain or loss of a network versus frequency. These sorts of plots are called Bode plots and are frequently shown in textbooks on circuit design and are produced by computer circuit analysis programs. A network analyzer consists of a swept frequency source that drives the network under test and two receivers. The first receiver is used to accurately measure the Reflection or input voltage to the network. The second receiver is called the Transmission channel and is used to measure the output of the network under test. The ratio of the output to the input level is displayed as dB and is the voltage gain or loss of the network. The source is swept over the frequency range of interest and a Bode response plot of the network results.

Commercial network analyzers range in cost from several thousand dollars on up and may be able to analyze circuits in the frequency range of milli hertz to microwave frequencies of 50 to 100 GHz. There are many special designs available that allow fast sweeps for automatic testing or very wide dynamic range for highly precise measurements.

 

The Personal Network Analyzer

The circuit presented here deviates from most commercial network analyzers in that the inputs are broadband. Commercial analyzers use very narrowband receiver inputs and at higher frequencies use superhetrodyne down conversion to convert the response at a lower frequency. The tradeoff of using wideband inputs makes the receivers respond the sum of all voltages input over the full bandwidth of the receiver. This lowers the dynamic range of the design and it also decreases the achievable accuracy. These tradeoffs were made because it lowers the parts count and total complexity of the design by at least 50%.

The personal network analyzer shown in figure 1 has the same basic blocks as a commercial analyzer. It has a swept source frequency, and two receivers. One receiver for the reflection signal and one for the transmission signal. The ratio of the transmission to the reflection signal is the response of the network under test. To program the analyzer and display the results, an RS232 communications link is used. This allows a program running on a PC to set the frequency, read the receiver inputs and plot the results directly on the PC's screen.

Receiver Circuit

The Reflection and Transmission receivers are identical in design and are based on the Philips NE604A IC. As shown in figure 2, the NE604A is used as a high dynamic range, wideband RMS to DC converter. The RMS voltage at the input of the IC is converted to a linear DC voltage at it's output by a series of limiting amplifiers. The limiting amplifier action serves to convert a 10 times change in input signal to a linear output voltage or about 44 mV output per dB change on the input (at pin 5 on the NE604A). Used by itself the NE604 responds to signals in the microvolt range to about 0.32 volts peak to peak.

To extend the dynamic range and increase the input impedance to the receiver, a buffer amplifier consisting of an Analog Devices AD847 is used ahead of the NE604A. A 10 K ohm resistor is used at the input of the buffer to set the input impedance of the receiver at 10 k Ohms, the AD847 buffers the input voltage to drive the lower impedance of the NE604A.

The AD847 is followed with a switchable 30 dB attenuator. The attenuator was chosen to be about 1/2 the available dynamic range of the receiver alone, thus extending the total system dynamic range significantly. The attenuator is controlled by the PC program operating the analyzer automatically to maximize the dynamic range of the receiver. With the attenuator switched in the NE604A does not overload with signals up to 10 volts peak to peak.

The AD847 was chosen because it has a wide bandwidth of 50 MHz and it's very low noise. The low noise is significant because it helps to keep the total noise floor down in this wideband design.

The RSSI output of the NE604A (pin 5) is buffered and filtered by a low frequency LM741 amplifier. The RSSI voltage is then sent to the microprocessor board for A/D conversion.

The one adjustment in this entire design is the gain adjust on the RSSI output voltage. Since the dynamic range of the NE604A is in excess of 80 dB in narrow band applications the RSSI voltage can swing from 0.2 volts to 4.8 volts. In this wideband design the noise floor is quite a bit higher (only about -60 dB). This makes the RSSI voltage range from about 1.5 to 4.8 volts.

The A/D converter used (on the microprocessor board) has an 8 bit, 0 to 5 volt input and if the RSSI voltage swings from 1.5 to 4.8 volts a reduced dynamic range is available. To make full use of the A/D converters input range, the RSSI voltage is offset negative by about 1.5 volts by the gain trimpot and resistive divider. This makes the RSSI range from about 0 to 3.3 volts. The 741 OPAMP is set for a gain of 1.5 to provide the A/D converter with a full 0 to 5 volt input swing.

The gain adjustment is used to set the full scale voltage on each receiver to be the same, thus allowing maximum dynamic range to be achieved. The ultimate resolution of the design is then approximately, 60 dB / 256 possible output codes of the A/D converter or about 0.23 dB per A/D converter LSB (Least Significant Bit) change.

DDS Source Circuit

The basic circuit of the DDS source built around the Harris Semiconductor HSP45102 has been around for several years [1]. This circuit is upgraded from the others in that I have added AC Coupling to the output along with three programmable outputs and a 16 MHz active low pass filter to control the harmonics some and flatten the frequency response.

The attenuators provide computer controlled outputs of 5, 2.8, 0.3 and 0.032 volts peak to peak. Having the source output switchable also helps increase the dynamic range of the analyzer by allowing the control program to reduce or increase the input voltage to the network under test to keep the receiver inputs in the active portion of their range. The attenuators are achieved by using simple shunt resistors to divide the approximately 160 ohm output impedance of the DAC to a lower voltage swing.

Controlling the sources absolute amplitude accuracy is not required in this design because the receivers operate ratiometrically. That is the ratio of the input to output is important, not the absolute value of the input voltage or output voltage.

The output of a digitally sampled sine wave has what is called a "Sine X over X" response. What this means is that as the output frequency approaches the clocked or sampled frequency the voltage rolls off in a sort of sine wave shaped response. The first null of the roll off is at the clock frequency. The low pass filter was optimized with the RF computer analysis program Touchstone [2] to provide slight peaking at the cuttoff frequency to compensate for the roll off. The resulting response is flat to within 1 dB from 10 Hz to 10 MHz and falls off to about -3 dB at 16 MHz.

The HSP45102 is a single chip circuit that when programmed with a 32 bit serial word will produce a digital equivalent of a sine wave on it's output. The output frequency is changeable by sending another 32 bit word to the device. The high speed version of the IC allows clocking at 40 MHz. Although the device operates with a 12 bit internal word size and has the capability to drive a 12 bit DAC, I have used (like all the others) the Harris CA3338. The reason is simple, this is a readily available, relatively low cost IC and it is capable of being clocked at 40 MHz also. 12 bit DAC's capable of operation at these frequencies will do a great job of lightening your wallet and the performance delta from going from 8 to 12 bits is not great enough to warrant the extra cost.

The DDS circuits that I have built have harmonics and spurs below -50 dBc up to 5 MHz and degrade to -40 dBc at 10 MHz. Above 10 MHz the alaising spurs come up to the -20 dBc level and start to appear below the programmed frequency. For example at 16 MHz output frequency, the largest aliased spur is actually around 5 MHz.

The master clock for the DDS is provided from a 40 MHz, CMOS oscillator. These "Canned" oscillators are very low cost and typically accurate to well within 0.01%. The clock is what sets the absolute accuracy of the output frequency. By using a 0.01% oscillator the output frequency will be within +/-1000 Hz when programmed at 10 MHz.

Microprocessor Circuit

The control microprocessor is the very popular PIC series from Microchip Technology (Chandler, AZ). It was chosen because it is a truly "Single Chip" computer and it's very low development and support cost. I built a complete development system including a programmer, first rate C Compiler [10] and UV Eraser for under $200.00.

The parts are readily available and when the reprogrammable types are used they can be erased and used over again in other projects.

One of the selling points of the 16C71 device used is that it is very easy to program the device for RS232 serial communication using only 3 wires to the host PC. The 16C71 also contains a four channel, 8 bit internal A/D converter.

The basic operation of the uP is to read commands from the RS232 port and set the source frequency, source and receiver attenuators and read the receiver output voltages (Via the PIC's built in A/D converter). A small C program was written to interpret ASCII commands from the RS232 line and set the appropriate bits high or low on the PIC's output pins. Full software handshaking is implemented by sending back an acknowledgment character ('*') after each successfully read command. If the synchronization between the PC and the PIC is broken for any reason the PC can then sense the error and resynchronize the programs.

When operating at 4 MHz the power dissipation is just a few milliamps and the RS232 connection can operate comfortably at a standard 9600 baud.

The internal, 4 channel A/D converter of the 16C71 device makes it easy to get data from the real world and send it along the RS232 into the PC for processing.

The availability of low cost, efficient C [10] and Basic compilers [11] for these devices also means that you don't need to learn yet another assembly language to put these processors to work. You can work in a comfortable high level language instead.

Of course these functions could have been implemented with half a dozen discrete logic chips and the result would have been the same. But the design and building time was cut from a week to just an afternoon using the PIC microprocessor.

Power Supply Circuit

The power supply is so straight forward that it doesn't hardly even need mentioning. A 25 volt center tapped transformer (available from Radio Shack) is used to provide unregulated +/- 17 volts to the three terminal regulators.

The only thing of special note here is that the regulator used for the +5 volt output should be the fairly accurate (+/-2% or better) type specified. This is because the regulator is used to power the microprocessor and the microprocessor uses this voltage as a reference for it's internal A/D converter. It is desirable to keep this voltage as close to 5 volts as possible to keep the dynamic range of the receivers optimized.

The +5 volt regulator should be mounted to a small heat sink as it dissipates some power due to the +17 volts at it's input.

Building The Network Analyzer

Most of the circuits are noncritical and can be built in just about any way you want. I built all of my circuits "Breadboard" style on small pieces of copper clad PCB. The copper clad provides a convenient ground plane and the circuits are built 3-Dimensionally above the board.

I built the source and receiver circuits on small pieces of copper clad that fit in small aluminum boxes of the type that Radio Shack sells. The small enclosures fully enclose the operating circuits and further help to reduce the noise floor.

The receiver circuits need some special care when building however. Separation from the input to the receiver to the input of the NE604 is a must if you are going to achieve the 60 dB dynamic range possible with this design. Because of the high impedance levels here (the NE604 has a 1.6 k input impedance) capacitive coupling is the mechanism that we need to protect from. I built my receivers pretty much like the schematic shows with the input circuit on top then the input signal folding back past the NE604 to the NE604's input. This created a coupling path from the middle section of the NE604 amplifier chain to the input. The net effect is regeneration, if not downright oscillation. The noticeable effect of regeneration is a noise floor only 30 or 40 dB down from full scale. To cure this problem I fashioned some shields from copper foil wrapped in insulating tape and placed them as shown on the schematic. The noise floor should drop considerably as they are positioned correctly in the circuit.

The DDS source circuit has just the opposite problem however. Because the impedance levels here are below 300 ohms, the coupling mechanism is magnetic. This means that circuit stray inductance is important. To keep the inductance to a minimum the wiring loop areas must be kept to the absolute minimum. This is achieved by keeping all lead lengths as short as possible and making short connections to the ground plane.

Because of the high speeds and large digital switching current on this board, decouple everything! The DDS chip needs to be decoupled to prevent ringing that will feedthrough to the DAC. The DAC needs to be decoupled to prevent digital noise from appearing at the output of the DAC.

The grounds from the DAC to the output active filter must also be as direct as possible. Otherwise any ground "Bounce" due to currents flowing through ground inductance will just show up directly at the output of the filter.

Each of the circuits ground plane's were tied to the shielding box with a piece of braid that is soldered to the copper clad ground plane and just slips between the box halves when assembled. The boxes themselves were then grounded together with more braid.

The PC Control Program

The PC control program called "Analyzer.EXE" is where the hardware gets all it's instruction as to how to perform. The basic sequence is as follows,

1) Setup for a sweep, get start and stop frequencies and number of points to be swept.

2) Get any other information about hardware external conditions that the hardware can't determine on it's own (i.e. is a X10 probe connected to an input?). Also set the min and max source amplitude limits if needed.

3) Start the sweep,

a) Send a frequency word to the source.

b) Autoscale source and receiver attenuators to keep the dynamic range optimized (if needed).

c) Make the receiver readings. Repeat if not settled.

d) Repeat for all frequencies in the sweep

4) Plot the sweep data (i.e. Make a Bode Plot).

5) Allow the user to view, measure with cursors, zoom around and print the resulting Bode Plot at his PC.

A continuously updating, single frequency "Manual Mode" is also available which allows peaking, nulling or tuning of the network under test in real time.

The program was written in 16 bit, Visual Basic for Windows and will run under Windows 3.1 and Windows '95. Visual Basic is perhaps the easiest way to program under Windows yet devised and allows easy graphics programming which is a large part of this program.

The program was written in three basic modules, the first is the low level hardware control. This is where the code actually communicates to the hardware via the RS232 to do such things as: Set the DDS frequency, make an A/D reading or switch an attenuator in or out.

The next layer is the high level hardware control where actual frequency sweep takes place and the autoscaling of the hardware attenuators takes place. The result of this layer is a data array of frequencies and dB ratio values for later display.

The last layer is the graphics display, the functions here control how the data is displayed, allow zooming in on the data and manipulate the on screen cursors.

Using a PC as a controller has the advantage of nearly unlimited number crunching ability. This comes in very handy when we want to convert nonlinear or slightly nonlinear functions to linear ones for display. The nonlinear function I'm referring to here is the RSSI output of the NE604 chip. The measured linearity over a 60 dB range is shown in figure 6, the bumps in the curve are about +/1.5 dB peak to peak.

Just because the chip has some nonlinearities doesn't mean that we have to live with them however. Using a precision set of attenuators I found the actual RSSI output for a known input over an 80 dB range of operation in 5 dB steps. I then wrote a function in the program that uses the measured data and linearity interpolates between points to improve the linearity to less than 0.4 dB.

As for speed the program is pretty much limited by the RS232 transmission time. On any computer faster than a 386, 33 MHz the RS232 time and not the computer speed will be the limiting factor. On my development system (a 50 MHz, 486) the program can do slightly better than 100 frequency points per minute.

Adjustments / Performance Checks

The only adjustment to be made on the analyzer is to adjust the RSSI full scale output. This is most conveniently done by using the PC control program in the "Manual Mode" and connecting the receiver inputs to the source output (use a 50 ohm termination on the source output). Set the source for a -20 dBv at 100 kHz output. Set the receiver attenuators to off. Then adjust the gain control (through the receiver shielding box) to be exactly 230 A/D counts on both receiver channels.

After this is done remove the cover from both receiver channels and measure the voltage at the bottom end of the 100 k ohm RSSI resistor. This voltage should be around -1.5 volts and the same for both channels. If it is much more negative then this suggests that the noise floor is too high. High noise floor is almost always caused by improper shielding around the NE604. Work on the shielding and make the gain adjustment again until the desired results are achieved.

Where do we go from here?

The usefulness and circuit insight gained by actually measuring ones designs is incredible. I for one learn much more by designing circuits and then measuring how they actually work than by doing a super analysis job up front. This is how I get a feel for building techniques and actual circuit parasitic's that affect performance.

The change of technology is rapid and I figure that in less than 5 years my next network analyzer will be much more precise. It is envisioned that single chip DDS sources will be available with frequencies approaching the low VHF range (100 MHz?). By using two of these sources a true tuned network analyzer could be built. This would entail using one DDS for the network source and using the second DDS source (programmed with a frequency offset) to drive receiver mixers. The receivers can then be made narrow band which will greatly expand the dynamic range by lowering the noise floor. The Linearity will also improve by having the RSSI circuit operate at a single frequency.

Having a narrow band IF also opens the possibility of adding a phase detector to the IF Limiter output stage. This would allow a true vector network analyzer to be built, and both Gain and Phase information could be displayed.

As the wireless revolution continues to drive IC performance, it is envisioned that log IF strip IC's will become available with much more precise and linear RSSI outputs, perhaps better than 0.1 dB linearity. The level of integration will continue to increase, allowing the next generation of network analyzers to be built with the same or even fewer components.

Tune back in again in about 5 years and see.....

References

[1] Hodgkinson, Bruce, "Julie Board - An Easy to build DDS Synthesizer", 73 Amateur Radio Today, August 1993.

[2] HP-EEsof, "Touchstone for Windows - Users Guide", 1995, Hewlett-Packard Company, Palo Alto, CA.

[3] Staff Article, "Direct Digital Synthesis - Part 2", Electronics + Wireless World, September 1992.

[4] Kushner, Laurence and Ainsworth, Marcus, "Spurious Reduction For Direct Digital Synthesis", Applied Microwaves and Wireless, Summer 1996.

[5] Hill, Allen and Surber, Jim, "Digital synthesis generates analog signals yet eases frequency hopping", Personal Engineering and Instrumentation News, August 1994.

[6] Williams, Fred, "A Microprocessor Controller For The Digital Frequency Synthesizer", QST, February 1985.

[7] Harris Corporation, HSP45102 Data Sheet, 1994, Harris Corporation, Melborne, FL.

[8] Harris Corporation, CA3338A Data Sheet, 1995, Harris Corporation, Melborne, FL.

[9] Harris Corporation, TB318 - The NCO As A Stable, Accurate Synthesizer, 1993, Harris Corporation, Melborne, FL.

[10] Custom Computing Services, "PCM - Pic C compiler reference manual", Custom Computing Services, Brookfield, WI.

[11] microEngineering Labs, "PBasic Compiler", microEngineering Labs, Colorado Springs, CO.

[12] ARRL, "Return Loss Bridges - 1996 Amateur Radio Handbook", American Radio Relay League,Newington, CT.

All commercial rights reserved.

Originally published in QST, February 1999 and copyrighted by the ARRL/QST.