BACKGROUNDIf you are involved in microwave operating in any form, you are familiar with being off frequency or not knowing your current frequency. We all have made numerous contacts by asking one person to send dashes while we hunted around on the band for them, only to fi nd them a few kilohertz or even tens of kilohertz away. Most of us have adapted to this way of operating. We have frequency counters and often know where we are likely to be even if we don’t know right where we are. SDR and softrock users can even plug in their presumed frequency in the software to get a relatively accurate readout on their computer if not on the IF radio. Many would argue that part of the fun of microwave operation is understanding what your LO is likely to be doing at any given time based on the temperature and environment and being able to compensate for this. And we all know that many contacts are made every year in spite of this condition.

No one would argue that it’s better to be on frequency. The source of frequency errors in most operations is the local oscillator (LO) present in all microwave transverters and its accuracy and stability problems. First, I’d like to discuss the differences in these two concepts (accuracy and stability), why they are important and what physical characteristics contribute to them.

STABLILITY Stability references the ability of an oscillator to remain at a given frequency given a changing environment. The environmental change could be factors such as temperature as well as other circuit-related factors such as voltage stability (regulation), and load on the oscillator itself. In general, when we look at oscillators that are built commercially, the most signifi cant factors in stability are the type of crystal chosen and the temperature variations to which the crystal is subjected. Power supply and load changes can be signifi cant, but are generally easy enough to compensate for that they fall way down on the list of real concerns. Temperature, on the other hand, is a signifi cant concern and in some operations (rover and portable) can have very large

swings throughout an operating session.

Stability of an oscillator is generally quoted as a function of the environmental condition (such as temperature). The anticipated delta in frequency is then specifi ed as parts-per-million or parts-per-billion. In other words, how many Hertz the frequency will change for each MHz in frequency of the oscillator in the case of parts-per-million. So for a 100MHz oscillator with a stability of 1ppm (also commonly quoted as 10-6), and a given operating temperature range of 0-70º C, the stability indicates that it could move up to 100Hz (1ppm x 100[MHz]). This doesn’t sound so bad until we use this oscillator as a reference source and multiply it up to 10 GHz. Now all of a sudden our 100Hz becomes 10kHz! This is why your 10 GHz transverter may wander around a few hundred Hz even when you just blow on it!

ACCURACYAccuracy of an oscillator talks about how likely the oscillator is to be on the designed frequency. So even if you have a high-stability oscillator that is not going to move much with the environment, it may not start out where you would like for it to start or it may move over a long period of time. Sometimes this is referred to as “long term stability,” but ultimately it amounts to an accuracy issue. Since most oscillators we use are based on quartz crystals, there are a number of factors related to the physical properties of quartz that will affect us. For example, most commercial oscillators will quote “crystal aging” in ppm. Sometimes this aging is specifi ed as a “fi rst year” number and then the maximum drift over the life span of the oscillator. So even though your oscillator may be on channel when you install it, after six months or a year it may have drifted off channel.

For us, the net effect of accuracy is that we need to periodically check oscillators against known references and adjust them to keep them on channel. How often you need to do this depends on the age of the crystal, how far off frequency you can afford to be and the band of operation.

A USB Programmable High Stability LOfor Microwave Transverters

Stephen Hicks, N5AC n5ac@n5ac.com

WHY DO WE CARE?Let’s talk about why we care about stability and accuracy. If an oscillator moves signifi cantly with a change in temperature, there are a number of issues that face us, as operators:

1. We spend more time chasing each other around the bands which can become signifi cant in some operating situations (contests, for one).

2. While roving along side another rover, I have seen similarly equipped stations to mine miss a contact because of a large separation in frequency. After the other person saw me make a contact and realized the other station was somewhere else in the band they were able to make the contact. I have no doubt that there are some contacts that do not get made simply because two stations cannot fi nd each other.

3. Without stability, some modes become impossible. For example, if you wanted to try your hand at terrestrial JT65, you need a certain amount of stability. If you don’t have it, it is simply not an option. Since signals are often weak to start with when operating JT65, it would be nice to know you are on the right frequency to start with!

4. As a simple matter of pride, it is nice to have the station radios on frequency

I’m sure there are other implications in other areas of operations such as EME.

What kind of Stability do we need?The required stability depends on what you are trying to do and the band of operation. If you would like to be stable on 2.3 GHz to around 500Hz across a temperature range, your oscillator needs to be in the 0.2ppm range (2x10-

7). The same stability on 10GHz requires an oscillator with a stability of around 0.05ppm (5x10-8). If your oscillator will be functioning in a narrower temperature range than is specifi ed by the manufacturer, you can also loosen these requirements some.

For running JT65, K2UYH suggests that initial

Maximum Frequency Error, Hz

Frequency, MHz

Stability, ppm 1296 2304 3456 5760 10368 24192 47088

5 6480 11520 17280 28800 51840 120960 235440

(10-6) 1 1296 2304 3456 5760 10368 24192 47088

0.5 648 1152 1728 2880 5184 12096 23544

(10-7) 0.1 130 230 346 576 1037 2419 4709

0.05 65 115 173 288 518 1210 2354

(10-8) 0.01 13 23 35 58 104 242 471

0.005 6 12 17 29 52 121 235

(10-9) 0.001 1 2 3 6 10 24 47

(10-10) 0.0001 0 0 0 1 1 2 5

Figure 1, Effects of LO stability by band

frequency error must be less than 600 Hz (accuracy) and during the QSO less than “a few Hz.” WSJT documentation from K1JT says stability must be better than 3Hz for the duration of the QSO. Below is a table showing LO stability vs. Band. In the intersection is the maximum drift over the temperature range. As a practical matter, though, an oscillator with a 1ppm temp stability promises to move less than 1ppm over the entire temperature range which is likely to be 0-70ºC. In reality, most of us will not begin a QSO with our LO frozen and expect it to stay put sitting on top of the PA. If you allow your LO to warm up before making a contact, these numbers are probably conservative by at a factor of 10 or so. In other words, if you leave your LO in a relatively temperature stable area, it is not going to drift as much as the manufacturer states. Even so, at 10GHz, we really need to be better than 10-8

to get under 10Hz drift for using some digital modes. Use the table below as a guide to help determine what sort of stability you need. I initially shaded the table to show green (or bottom) areas as less than 500 Hz, and

the read areas (top) as greater than 3kHz, thinking that if you were within 3kHz, you would be relatively easy to fi nd and that for 500Hz or less, no tuning would be required.

How Do We Get There?So what does it take to get to these levels of stability? Figure 2 shows some common solutions and their stability. I’d like to take a moment to discuss the terminology which most readers will already know, but

some may not.

To offset the natural tendency of a crystal to move in frequency with temperature, we can work to hold the temperature constant or counter its effects on the crystal. A TCXO is a Temperature Compensated Xtal (crystal) Oscillator. The compensation in a TCXO is generally a resistor and capacitor compensation network that counters the crystal’s temperature sensitive nature. An OCXO is ovenized and so includes a heating unit and temperature sensitive components to turn the oven on and off to attempt to hold the crystal at a constant temperature. An OCXO is about the best frequency stability we can get out of a crystal. After this, we need to move to more stable references, generally based on hyperfi ne transitions in the ground states of specifi c atoms. The NIST uses Cesium particle fountains for standards and the tubes alone cost upwards of $35,000. Rubidium has since replaced Cesium in most commercial applications as a less expensive alternative.

In a December 2007 article in WIRED magazine1, Quinn Norton suggests that there is an analog to Moore’s Law going on in timekeeping or reference devices—that they are improving by a factor of 10 every decade. He states that the NIST-F1 using cesium with a resonant frequency of 9.1GHz may be replaced in the future with a Calcium (resonant frequency of 456THz) or Ytterbium (518THz) for greater accuracies. Indeed the NIST is currently developing a chip-scale atomic clock2 that could replace our use of crystals in oscillators. Whether you believe these predictions for the future or not, there are advantages of having a stable LO.

Shopping for StabilityAs I have gone shopping for stable oscillators for my own projects, I have found that fi nding an oscillator on the required frequency (or getting it re-crystalled) and getting the right stability for projects can be a project in itself. There are a number of stable references that are

Type of device Typical Stability Range Typical Stability, ppm

Uncompensated Crystal 2-5 x 10-5 25

Uncompensated VCXO 5 - 100 x 10-6 15

TCXO 0.2 – 5 x 10-7 1

OCXO 1-50 x 10-8 0.05

HP 58540A GPS Disciplined OCXO 1 x 10-11 0.00001

GPS Disciplined Rb Osc 1 x 10-12 0.0000002

NIST-F1 Cesium Atomic Clock 5 x 10-16 0.00000000005

Figure 2, Types of oscillators and their relative stability

used in test setups and commercial telecommunications, though. Many amateurs are picking up Z3801’s and other 10MHz references originally intended for these type of applications. I wondered: is there a way to easily use these references with their higher stabilities

as a reference for our LO’s? I got to thinking about cellular phones which need stable oscillators that can handle transmission of high-speed digital data in the 2GHz range and decided that there must be some good ICs available for solving this problem for the cellular and communications industry.

I looked at a number of alternatives, but didn’t particularly want to roll my own PLL or VCO if I didn’t have to. I uncovered a synthesizer chip designed for use in cellular phones and similar applications at Silicon Labs.3 The Si4133 is an IF/RF synthesizer with an integrated VCO. It requires a stable reference and produces IF & RF LO frequencies out the other side of the part. The RF VCO is tunable to 5% around a center frequency and since the part has two RF synthesizers (for dual-band phones), I could use these to obtain around a 10% theoretical tuning range.

Since I have been using exclusively DEMI transverters, I decided my fi rst project would be to create a replacement LO for 2, 3, 5, and 10 GHz transverters that could use this part. All of these LO’s start with a ~200MHz crystal and exit the MICROLO at around 1GHz. They are then further multiplied on the main board of the transverter. Since all of these LO’s are in the 1GHz range, I decided that the best thing to do was to replace this part of the LO, leaving the remainder of the LO multiplier chain intact in the transverter. I fi gured I would get a very stable oscillator (somewhere in the 0.01ppm range) and put it on the board along

with the Si4133 and would have a much more stable LO. So I went shopping for a TCXO/OCXO.

The Si4133 needs a reference in the 2-26MHz range. Due to the mechanics of the VCO loop, the frequencies we are going to synthesize and the availability of external references at 10MHz, I chose 10MHz as the reference frequency. I anticipated that I would be able to acquire a reference source from the standard distributors (Mouser, DigiKey, Newark) for $10-20 that would do what I want. Boy was I wrong! As I surveyed the landscape, I found that high-stability TCXO and OCXO’s are a niche market and mostly handled through distributors other than the off-the-shelf variety. I looked at a number of companies in addition to the distributors including: Connor-Winfi eld, Valpey-Fisher, Raltron, Crystek, Kyocera, Isotemp, ECS, Mercury, NDK, MtronPTI and Wenzel. What I found is that each had a solution for me, but most all are “custom” solutions that would be manufactured as soon as I sent in my purchase order for 1,000 units. Actually it’s not quite this bad, but often an order for a few tens of units costs as much as a few hundred units due to the startup costs, etc. Further, the costs were much higher than I expected. A typical TCXO was in the range of $50 and an OCXO was in the $150-350 range. This was clearly not going to be adequate for what I was trying to do—to change out the LO in a $400 transverter with a $500 LO seemed to be a bit of overkill.

I came to the realization that two solutions would really be best: First, an oscillator that could use an inexpensive TCXO and get an order of magnitude improvement in stability at a sub-$100 cost (for the entire LO). This LO could be used to replace existing LO’s and yield a higher level of stability. Second, I wanted an upgrade path for this LO where a higher stability oscillator could be used to drive multiple transverters. So to start, the replacement LO would be able to provide better stability than existing LOs at a comparable price with an upgrade path of just adding an external reference.

ProgrammabilityToday, most LO’s are set by a specifi c crystal and changing the frequency of the LO is impractical once everything is constructed and tuned. Since the Si4133

is programmable inside a range, it is possible to build a “universal LO” that could be set to any frequency in a given range. Fortunately, all of the DEMI transverters from 1296 through 10GHz need LO frequencies in the range 1080 to 1152 (before any multiplier chain) and these all fi t nicely in the range of the Si4133. The Si4133 needs to have frequency data loaded into internal registers via a microprocessor. One option would be to load this data via a dip switch connected to the microprocessor, but since most amateurs have ready access to a personal computer, I opted to put a USB port on the board and select a microprocessor capable of USB operation. I wanted for this LO to be easy to setup and to change later if needed.

There are a number of USB solutions available including stand-alone USB-to-serial solutions, but to keep cost down I selected a Microchip PIC with built-in USB. Microchip makes four devices that have a built-in USB interface that required no external parts save a conenctor. These parts are in the PIC 18F4550 family.

Si4133 Internals The Si4133 is a dual-band RF synthesizer with integrated VCOs. A block diagram of the Si4133 is shown in fi gure 3. The two RF synthesizers are each tuned to a center frequency via an external inductor. Then the data loaded into the part selects the operation frequency. Feedback to the microprocessor allows the processor to know when the frequency selected is near or at the edge of the internal VCO range of the part. As you can see in the block diagram, only one of the RF synthesizers exits the part at a time, which works well for this application since we need only one

Rev. 1.6 6/06 Copyright © 2006 by Silicon Laboratories Si4133

Si4133Si4123/22/13/12

DUAL-BAND RF SYNTHESIZER WITH INTEGRATED VCOSFOR WIRELESS COMMUNICATIONS

FEATURES

Applications

Description

The Si4133 is a monolithic integrated circuit that performs both IF and dual-band RF synthesis for wireless communications applications. The Si4133includes three VCOs, loop filters, reference and VCO dividers, and phasedetectors. Divider and powerdown settings are programmable with a three-wire serial interface.

Functional Block Diagram

� Dual-band RF synthesizers��RF1: 900 MHz to 1.8 GHz��RF2: 750 MHz to 1.5 GHz

� IF synthesizer ��IF: 62.5 to 1000 MHz

� Integrated VCOs, loop filters, varactors, and resonators

� Minimal (2) number of external components required

� Low phase noise � Programmable powerdown modes� 1 µA standby current� 18 mA typical supply current � 2.7 to 3.6 V operation� Packages: 24-pin TSSOP,

28-lead QFN��Lead-free and RoHS compliant

� Dual-band communications� Digital cellular telephones GSM 850, E-GSM 900, DCS 1800,

PCS 1900� Digital cordless phones� Analog cordless phones� Wireless local loop

IFOUT

IFLA

IFLB

RFOUT

XIN

PWDN

SDATA

SCLK

SEN

IF

RF2

RF1PowerdownControl

ReferenceAmplifier

SerialInterface

AUXOUT IFDIV

�R

�R

�R

�N

�N

�N

PhaseDetector

22-bitData

Register

TestMux

RFLC

RFLD

RFLA

RFLB

PhaseDetector

PhaseDetector

Patents pending

Ordering Information:See page 31.

Pin Assignments

Si4133-GT

Si4133-GM

1 24

2 23

3 22

4 21

5 20

6 19

7 18

8 17

9 16

10 15

11 14

12 13

SCLK

SDATA

GNDR

RFLD

RFLC

RFLB

GNDR

RFLA

GNDR

GNDR

RFOUT

VDDR

SEN

VDDI

IFOUT

GNDI

IFLB

IFLA

GNDD

VDDD

GNDD

XIN

PWDN

AUXOUT

GNDPad

SCLK

SDAT

A

GN

DR

RFLD

RFLC

RFLB

GNDR

RFLA

RFO

UT

VDD

RSE

N

VDD

I

IFO

UT

GNDI

IFLB

IFLA

GN

DD

VDDD

GNDD

XIN

PW

DN

AUXO

UT

21

20

19

18

17

16

15

8 9 10 11 12 13 14

28 27 26 25 24 23 22

1

2

3

4

5

6

7

GN

DR

GNDR

GNDR GNDD

GN

DI

GN

DR

Figure 3, Si4133 Block Diagram

RF frequency synthesized and having an internal mux allows two different ranges to be selected by the LO if required. Incidentally, the IF synthesizer has a range (62.5-1000MHz) that works nicely for some of the lower bands including 902 and 222 also.

The serial interface to the part uses standard SPI-style serial data input although the data sent to the part is a somewhat unconventional 4-bit address and 18-bit data streams.

Fortunately, the part is available in a human-solderable TSSOP package in addition to the QFN package.

The Si4133 output power is specifi ed in the -7dBm range which is too low for use directly as an LO in our transverter applications so an amplifi er needed to be added between the synthesizer and the mixer (or multiplier chain) of the transverter. Al Ward, W5LUA, suggested an ABA-54563 which has around 20dB gain at 1GHz and a P1dB of +16dBm so after some help from Al on layout and component selection for this part, the LO’s output ranges from around -10dBm to +10dBm.

Design DetailsThe schematics for the LO are included at the end of this paper. The power supply is a little more complicated than normal just because the LO needs to be able to run off of either a computer USB port or off of the supply inside of the transverter. There are two LEDs on the board to indicate that the LO is powered, one for USB power (Red) and the other for Transverter power (Green). There is a third LED (blue) to indicate when the synthesizer is locked. At the time of this writing, I have not completed the software, but the idea is that the LED will be solid for a locked condition, fl ashing for locked, but near VCO edge and off for unlocked.

There is also layout room to put a couple of different types of TCXO parts directly on the board. Again, if you do not need very high stability, a TCXO can be placed directly on the board. If a higher stability is required, an external reference can be piped into the LO through the supplied input port on the board.

Phase NoiseAs an LO is multiplied up, the phase noise of the oscillator also gets multiplied up at 20dB/decade of multiplication. So a -60dBc/Hz phase noise at 100Hz for a 1GHz LO source oscillator becomes -40dBc/Hz for a 10GHz LO. The practical question about phase noise in an LO are:

1. Is the phase noise audible in your receiver?2. Is the phase noise signifi cant enough to QRM your

neighbors if you run your TX signal through a HPA?

3. Will near-in phase noise prevent some digital modes from operating due to the distortion caused by phase noise.

I’d like to say that I have the answers to all of these questions, but I have not yet had an opportunity to do all of the resting the is required to answer these questions. This is a good project for the upcoming year.

While I was off laying out PCBs and writing software, Al Ward volunteered to do some phase noise measurements on the Si4133 with different reference oscillators attached. The preliminary measurements using the Si4133 connected to a DEMI multiplier chain

Si4133

Rev. 1.6 13

Figure 9. Typical RF2 Phase Noise at 1.2 GHzwith 200 kHz Phase Detector Update Frequency

Figure 10. Typical RF2 Spurious Response at 1.2 GHzwith 200 kHz Phase Detector Update Frequency

102

103

104

105

106

−140

−130

−120

−110

−100

−90

−80

−70

−60

Offset Frequency (Hz)

Pha

se N

oise

(dB

c/H

z)

were comparable to those of a Frequency West Brick. I should have some more detail on this by conference time. The Si4133 data sheet has some phase noise plots run by Silicon Labs on the part at 1.2 GHz shown in Figure 9 above that should also provide some insight into what the part is capable of. The phase detector update frequency also has a marked impact on phase noise so for most frequencies I decided to use a higer update frequency.

Software NotesI’m in the process of working through the details on the fi rmware for the PIC as well as the PC software to communicate with the PIC. Since the synthesizer powers up “dumb,” meaning it does not have any non-volatile storage to tell it what frequency it should be on, the microcontroller is required to load the Si4133’s internal registers with the parameters to get the desired output frequency (this part was not really intended to be a stand-alone LO, but part of a larger system such as a cellphone). From an end-users’ perspective, the board powers up with the correct frequency because the microprocessor looks up the previous settings from its non-volatile memory (fl ash) and reprograms the Si4133 at power-up. All of this is really the “easy part” and the hard part is getting all of the USB software to work correctly.

A USB peripheral needs to be able to respond to numerous requests from the PC for everything from I/O to questions about how much power it is consuming and from what (the PC or another power source).

I decided to make the PIC software fairly “stupid” in respect to the operation of the Si4133. Sepcifi cally, I want to use it to load the desired values into the registers in the Si4133 and respond to the USB, but not perform any real calculations. This is primarily because the PC software is signifi cantly easier to update over the web, etc. without having to do any real programming on the PIC.

SUMMARYThe objective of the project was to create a more stable LO that would help get more stations on frequency. While all the data is not “in” since the software is

not complete, this LO appears able to produce a stable carrier that is proportional to the stability of the reference used with minimal phase noise. Above are actual size PCB top and bottom layouts for the board. With any luck, I’ll have the LO working and ready for a demonstration by the conference.

(Endnotes)1 December 12, 2007 WIRED Magazine: http://www.wired.com/science/discoveries/news/2007/12/time_nist?currentPage=all2 Chip-Scale Atomic Devices at NIST: http://tf.nist.gov/ofm/smallclock/index.htm3 Silicon Labs: http://www.silabs.com