Miles, Hosking: A Versatile Hybrid Synthesizer… Page 1 of 47
Authors’ draft of article published in
March/April 2004 QEX/Communications Quarterly
Please refer to http://www.qsl.net/ke5fx/synth.html for the latest updates and errata
A Versatile Hybrid Synthesizer for UHF/Microwave Projects
John Miles KE5FX
Richard Hosking VK6BRO
Ever want to build a “DC to daylight” receiver or digitally-controlled spectrum analyzer?
How about a transceiver for the Amateur UHF bands, or a signal generator with octave-band coverage? At the heart of each of these projects is a local oscillator with good stability and spectral purity. Our goal in this article is to present a versatile and practical synthesizer design
that can address almost any homebrewer‟s need for a digitally-tunable signal source from VHF to
Unlike most published approaches, we‟ve focused our design efforts as much on affordability, flexibility and reproducibility as on “specsmanship.” The synthesizer we‟ll describe offers continuous coverage between 1 and 2 GHz with fast switching time, sub-1 Hz tuning resoluion and very good close-in noise performance. There‟s nothing to tweak or align with exotic test equipment – if you build it, it will work. Best of all, every component is available off-the-shelf from Mini-Circuits, Digi-Key, and Analog Devices for a total of less than US $200.00!
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 2 of 47
When it comes to modern synthesizer design trends, numerous authors have made the case for a hybrid topology that takes advantage of the combined strengths of direct digital
1,2,3,4,5synthesis (DDS) and traditional phase-locked loop technology. Before these hybrid
techniques became practical, PLL synthesis often involved awkward tradeoffs between frequency step size, freedom from spurious outputs, and overall loop complexity. While standalone DDS chips do not yet have the output frequency range (and in many cases, the spectral purity) to offer an across-the-board replacement for PLL technology, it‟s easy to use DDS technology to build a PLL synthesizer with arbitrary frequency-control precision, competitive spectral purity, and low overall complexity.
In its most basic form, a hybrid synthesizer uses a DDS source to provide a stable, clean, and precisely-tunable reference for a conventional PLL. The output of a hybrid synthesizer derives its tuning precision and stability from its DDS reference while providing the frequency coverage range typical of a PLL. The approach we‟ve taken is similar to that in Cornell Drentea,
6KW7CD‟s recent microwave DDS-driven PLL article, adapted to use newer, more readily-
available parts and improved post-DDS filtering. Tuning control is provided via the PC parallel port or an Atmel microcontroller. Like Cornell‟s design, our synthesizer can be used as a standalone VHF/UHF/microwave source or with an external frequency divider to achieve exceptional noise performance in HF applications.
The synthesizer‟s heart is the AD9852 DDS chip from Analog Devices. The AD9852 is used to generate a finely-tunable reference signal near 10.7 MHz for the PLL. Chosen for its superior performance over the less-costly (and somewhat more popular) 10-bit AD9850, the 12-bit AD9852 is clocked by either an onboard 10 MHz oscillator or an external 10 MHz source, using the chip‟s clock-multiplier feature to generate internal clock frequencies between 80 and 120 MHz.
After passing through a crystal filter to tame any wideband spurs, the signal from the DDS is amplified and converted to a square wave by an LT1016 comparator from Linear
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 3 of 47
Technologies. The filtered and squared PLL reference signal from the comparator, along with a sample of the signal from the synthesizer‟s VCO obtained from a resistive divider, is applied to any of several programmable PLL synthesizer ICs from National Semiconductor‟s LMX2306/16/26 or Analog Devices‟ ADF4110/11/12/13 family. These PLL chips are marketed
toward the wireless- and cellular-communications industry, but their low prices, ease of use, and availability in small quantities make them attractive for Amateur Radio applications from the lower VHF region to frequencies well beyond the 2.4 GHz band.
Generation of the synthesizer‟s output signal is handled by a commercial varactor-tuned
voltage-controlled oscillator module. A wide variety of prepackaged VCOs are available with industry-standard PC board footprints from vendors such as Mini-Circuits and Synergy Microwave, offering various tuning ranges, output power levels, harmonic and noise performance specifications, and supply voltages. The synthesizer project described here uses the ROS-
2150VW from Mini-Circuits, which provides an impressive 970-2150 MHz tuning range for about $30.
Overall loop characteristics such as phase margin, bandwidth, and lockup time are determined by a third-order active loop filter based on the low-noise OPA27 opamp from Texas
Instruments. The opamp not only filters the digital signal from the PLL chip‟s charge-pump output,
but also amplifies it to the 0.5 - 25V level required to make the most of the VCO‟s tuning range.
Following the VCO and resistive divider, a GALI-5 MMIC from Mini-Circuits provides
approximately 16 dB of gain. With the ROS-2150VW VCO and divider as shown, power available at the output jack at mid-band is approximately +14 dBm with less than +/- 2 dB of variation between 1000 and 1800 MHz.
A key specification of any PLL frequency synthesizer is the spectral purity of its output signal. Phase noise, also known as “jitter,” is caused by random short-term excursions in the
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 4 of 47
carrier‟s phase and can arise from a variety of causes. Amplitude noise is the other aspect of
composite noise performance as observed with a receiver or spectrum analyzer, but it is usually insignificant compared to phase/frequency jitter. Finally, in addition to composite AM/PM noise, the synthesizer‟s output signal may also contain discrete spurs that appear as sidebands on either or both sides of the carrier. For a detailed discussion of the nuances of noise and spur performance, see Dean Banerjee‟s outstanding PLL Performance, Simulation, and Design,
downloadable at http://www.national.com/appinfo/wireless/files/DeansBook_4_01.pdf. (A printed and bound version of Banerjee‟s book is also available at http://www.amazon.com.)
Unlike a simple LC oscillator stage, the factors that can contribute to noisy or otherwise-impure signals at the output of a PLL are almost too numerous to count. Worse, it‟s impossible to label a given synthesizer with a simple “spectral purity” specification that will allow direct comparisons with its peers. Noise and spurs are two-dimensional quantities – to evaluate their
severity, we must consider both the amplitude relative to the carrier signal and the frequency
offset from the carrier at which the effect in question was measured. Any synthesizer which will be used in a receiver or transmitter must be carefully designed to minimize noise and spurs at frequency offsets which have the potential to degrade the receiver or transmitter’s performance.
The last point is significant - a synthesizer serving as a wideband FM receiver‟s local oscillator can be designed to much looser specifications than one intended for use in a 40-meter contest-grade rig.
A detailed discussion of noise causes and cures is beyond the scope of this article. However, in a nutshell, PLL noise and spurious signal analysis must consider reference noise,
VCO noise, and the multiplication effect of the loop. The design of the loop filter is critical, as
the bandwidth of the loop should be tailored to the VCO‟s needs. It is also important to consider extraneous noise-and-spur sources such as power supplies and intermodulation effects
between various parts of the circuit.
Loop Multiplication Effect
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 5 of 47
Any phase-locked loop will increase the phase-noise amplitude of its reference source by 20 * log(N) dB within the loop bandwidth, where N is the loop‟s overall frequency-multiplication
factor. Any discrete reference spurs within the loop bandwidth will also be amplified by a similar factor, noting that their frequency offset from the output carrier will remain the same as their offset from the reference. Consider a synthesizer with a 1-GHz output frequency and a 3-kHz loop bandwidth, whose reference frequency is 1 MHz with discrete sidebands at +/- 1 KHz and –80
dBc. N is 1000 in this case, and 20 * log(N) is 60 dB. At its output, the synthesizer will show
sidebands at +/- 1 KHz and –20 dBc, which may not be acceptable for many applications. It is
therefore critical to use a high-quality reference source for microwave PLLs with their high N
ratios -- or, failing that, a very narrow loop bandwidth with a high-quality VCO.
Reference-Related Noise and Spurs
The random phase-noise performance of a DDS is quite good. It is determined by either the noise of its own reference - typically a very clean crystal oscillator - or the noise-floor limitations of the VLSI process by which the DDS chip was fabricated (typically –140 dBc/Hz for
ECL technology and –150 dBc/Hz or better for modern CMOS parts such as the AD9852).
However, while its phase-noise performance is adequate for most purposes, the output of a DDS exhibits discrete sideband spurs due to phase truncation and timing tolerances in the lookup table and DAC. These are often significant enough to rule out the use of a standalone DDS as the first local oscillator in a high-performance HF receiver. Discrete spurs in the DDS reference can appear in the synthesizer‟s output signal at significant carrier offsets, falling away outside the loop
bandwidth after being amplified by 20 * log(N) dB within the loop bandwidth (see above).
In reality, our loop‟s vulnerability to spurs in its reference source is greater than one might expect. Both the National Semiconductor LMX2326 and Analog Devices ADF4112 PLL chips exhibit a readily-observed (yet apparently undocumented) tendency to respond to reference-signal spurs at offset intervals corresponding to their internal comparison frequency, from DC to 100 MHz and beyond. Consider a loop with a 1 MHz comparison frequency, achieved by programming the PLL chip for a reference-divider modulus (R) of 10 with a DDS-generated
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 6 of 47
reference signal at 10 MHz. As expected, DDS spurs close to the 10 MHz reference will appear in the synthesizer‟s output signal as noted above. Additionally, any DDS spurs appearing near 1 MHz intervals on either side of the 10 MHz reference frequency will appear in the output signal exactly as if they had been generated in the vicinity of 10 MHz. For example, a spur at 7.001 MHz
would produce sidebands at +/- 1 KHz from the carrier at the output of the synthesizer. This effect has been noted with the National chip‟s reference/evaluation board as well as our prototype ADF4112-based synthesizer. It is clear that low-pass filtering the DDS reference is not enough –
we must bandpass-filter it to suppress as many spurs as possible across the entire RF spectrum.
Our synthesizer addresses DDS spur problems by severely band-limiting the loop‟s
reference signal with an inexpensive 4-pole monolithic crystal filter. With this 15-kHz-wide filter in place, a series of 5,000 automated measurements taken at randomly-selected frequencies between 1000 and 1800 MHz revealed no significant spurious responses at any frequency. Without the filter, the synthesizer‟s overall spur performance was dramatically worse. In the latter test, almost every randomly chosen frequency exhibited at least one noticeable spur.
The Loop Filter and VCO
The loop filter in a PLL is designed to attenuate high-frequency components of the loop error signal so that they do not modulate the VCO output. In practice, the loop filter cutoff is commonly set at between 2 and 5% of the reference frequency to obtain adequate attenuation of comparison-frequency sidebands appearing at the phase detector‟s output. In our circuit the comparison frequency is approximately 900 kHz, so this issue can be largely ignored. We consequently design the loop filter for best system phase-noise performance based on the VCO‟s
own noise characteristics.
At offsets from the VCO carrier less than the loop filter bandwidth, the PLL attenuates noise contributed by the VCO. At offsets that are significantly greater than the PLL‟s loop bandwidth, the principal source of noise is the VCO itself. Within the loop bandwidth, the noise performance is determined by the PLL reference source and the phase detector, and possibly the broadband noise floor of the frequency dividers as well. Even the noisiest VCO can be cleaned
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 7 of 47
up by a PLL within its loop bandwidth, but extremely wide loop bandwidths carry a price of their
ADF4110/11/12/13, the inband noise own. As described in Analog Devices‟ data sheet for the
floor of a typical inexpensive phase-frequency detector is limited to –85 to –90 dBc/Hz, after
allowing for the multiplication effect of the loop as mentioned above. The phase-noise specifications of a good-quality VCO at offsets greater than 10 kHz from the carrier will be better than this -- typically around –95 to –110 dBc/Hz. Consequently, it makes sense to choose a loop
bandwidth narrower than the carrier-offset points where the VCO‟s free-running noise profile
crosses the PLL‟s own noise floor.
Besides requiring physically larger components, a narrower-than-necessary loop bandwidth allows more VCO noise to appear in the output signal. It also slows the loop lock time. The latter may be an issue where rapid tuning is required, such as in a sweep generator or spectrum analyzer application. Conversely, an excessively-wide loop bandwidth means that the PLL is actually contributing noise to the output, degrading the VCO‟s noise profile rather than improving it.
With the Mini-Circuits VCO family we‟ve specified, loop bandwidths in the 1.5 kHz – 2.5
kHz range seem to yield the best compromise between component size, lock time, and noise performance. Some typical composite-noise results are shown in Figure 4. For comparison purposes, the red trace was obtained from the first local-oscillator section of an Icom IC-R7000 VHF/UHF communications receiver. The R7000‟s synthesizer splits its 770-1290 MHz output
range between two separate VCOs, using a narrow loop bandwidth to suppress the reference-frequency spurs at 5 kHz. As a result, its close-in noise performance suffers relative to the hybrid synthesizer, but at offsets greater than a few kilohertz from the carrier, the Icom‟s dual discrete VCOs demonstrate their superiority.
There‟s no escaping the fundamental truth: a high-quality synthesizer design has to start with a high-quality VCO. As the Icom design shows, multiple oscillators with narrow tuning ranges perform better than octave-band units like the ROS-2150VW, assuming the basic tank-circuit technology remains the same. Nevertheless, the Mini-Circuits parts hold up surprisingly
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 8 of 47
well when compared to discrete-component VCO topologies like the IC-R7000‟s that are beyond
the reach of most homebrewers.
Phase-Detector Noise and Comparison Frequency
Apart from the multiplication factor described above, noise contributed by the phase detector increases with increasing comparison frequency (the frequency at which the phase detector itself operates, after any prescaling and division is accounted for). Further, in our design, the higher the comparison frequency the wider the frequency range the DDS reference must cover, potentially making post-DDS filtering more difficult with off-the-shelf parts. Using a cheap 10.7 MHz FM crystal filter with 15 KHz bandwidth, we found a PLL reference frequency of about 1 MHz (10.7 MHz divided by 11) gave the best tradeoff between phase detector noise and DDS bandwidth.
Another potential source of discrete spurs in the synthesizer output is crosstalk between the DDS and PLL chips. The AD9852 DDS is a power-hungry device capable of radiating and conducting high-amplitude RF onto circuit-board traces in its vicinity. Without extensive physical shielding between the two components – e.g., placing them in separate RF-tight enclosures -- the
PLL output exhibits spurs when tuned near a multiple of either 1/2 or 1/3 of the DDS chip‟s master clock frequency.
If the DDS is driven by a 100 MHz clock, for example, spurs can be observed on either side of the synthesizer‟s output signal when tuned near 1033 and 1050 MHz as well as most other multiples of 33 and 50 MHz. One spur appears at the fractional clock-frequency multiple in question, while an identical spur appears at the same offset on the other side of the output signal. Like other reference-derived spurs, these spurs begin to fall off in amplitude once their offset from the output frequency exceeds the PLL‟s loop bandwidth.
At a few of these problematic output frequencies, the crosstalk effect is severe enough to destabilize the loop and cause intermittent oscillation. The solution we‟ve implemented takes
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 9 of 47
advantage of the AD9852‟s built-in clock multiplier feature. Instead of clocking the DDS at a constant 100 MHz, we apply 10 MHz to the DDS‟s clock input and use its clock multiplier to select
one of five possible clock frequencies between 80 and 120 MHz, maximizing the distance between any harmonic of fDDSClk/2 or fDDSClk/3 and the synthesizer‟s output frequency. This technique keeps the nearest fractional clock harmonic over 1.5 MHz away from the carrier at any given frequency, eliminating the problem entirely except for residual leakage of clock harmonics into the signal path associated with U204‟s input and output. Even without any additional
shielding on the board, the latter spurs are seldom worse than –80 dBc.
While this approach avoids spurs due to this intermodulation effect, it does carry the penalty of increased software complexity. The extra calculations pose no significant burden to a PC or high-performance Atmel AVR controller, but may be a consideration if a less-capable microcontroller is used to drive the board.
Power supplies and noise
Some sections of the circuit are very sensitive to power supply noise. In particular, overall phase noise performance will be degraded if supplies to the VCO and the PLL chip are not adequately filtered and decoupled. Typical IC voltage regulators are very convenient to use, but they may exhibit wideband noise at magnitudes many times greater than a well-designed discrete-component regulator. We used Zener diodes as relatively-quiet references for critical parts of the circuit. Additionally, separate regulators are used to isolate different parts of the circuit, notably the digital and analog DDS sections.
By toggling the synthesizer between its frequency extremes while monitoring the VCO tuning port with a digital oscilloscope at 5 milliseconds/division, the loop‟s lockup time and damping characteristics can be observed directly. As the graphs in Figure 5 show, worst-case lockup time is approximately 25 milliseconds. Like the phase-noise graphs, the lock-time graphs
Miles, Hosking: A Versatile Hybrid Synthesizer… Page 10 of 47
were taken with an ROS-2150VW VCO, with the results compared to those obtained from the Icom IC-R7000‟s first local oscillator. The hybrid synthesizer‟s loop bandwidth was approximately 2 kHz.
Assembly and Operation
To obtain acceptable performance at microwave frequencies, it‟s necessary to use small components. The ICs used in the synthesizer come in TQFP, SOIC and SSOP packages with pin spacing as small as 0.6mm. It is nearly impossible to work with these devices without a printed circuit board. While our prototype was constructed dead-bug style on a bare copper-clad board, the author‟s eyesight and nerves have yet to recover from the experience!
Many hobbyists are intimidated by the precision and small dimensions involved with surface mount construction, but the truth is that SMT homebrewing is relatively easy to master with inexpensive tools, a modicum of patience, and a clever trick or two. In fact, surface-mount construction is a boon in disguise: because all of the components and their pins are accessible from the top of the board, a dual-layer surface-mount board carries almost all of the „tweakability‟
advantages of dead-bug or Manhattan construction, while surface-mount components tend to be easier to purchase and stock due to industry standardization. Conversely, traditional DIP IC packages are fading from the scene at a frightening pace. Most modern RF ICs including DDS and PLL parts are simply not available in DIPs.
Even if this is your first surface-mount project, you‟ll find that assembly goes smoothly
with a few basic tools you may already own. A grounded soldering iron with a clean, fine tip is a must, and many constructors will find a lighted magnifier helpful as well. Manipulating the 1206-sized resistors and other smaller components is very difficult by hand, but a pair of cosmetology-grade tweezers such as the 3 3/4" Rubis model
(http://www.folica.com/removal/rubis_costwe132.htm) makes the job trivial. Don‟t settle for whatever‟s on sale for $1.99 at your local drugstore – when you‟re doing SMT work, a good pair