The ADF 5355 is a versatile fractional N synthesiser that covers the frequency range 54 MHz to 13.6 Ghz . It uses and on-chip VCO arrangement and is capable of milliHertz frequency resolution at 10 GHz. They are readily available from Chinese vendors on the usual auction sites incorporated into basic development boards. The cost of these boards is comparable with the cost of the chips in one off quantities. For some time I thought that a useful signal generator could be produced based on one of these boards.
In the past I have used them for low powered personal beacons and as local oscillators and signal sources all programmed with Arduinos. Recently I came across the Arduino based signal generator design by Christian Petersen DD7LP based on the ADF4351  and I used his Arduino “sketch” as the starting point for my project. The ADF5355 is capable of operating with frequency steps of milliHertz at 13 GHz, and to have the required arithmetic precision for one Hertz resolution was going to be beyond the capabilities of an 8-bit Arduino ATmega328P. It also has a more complex register structure than the ADF4351. Fortunately there are a number of much more powerful processors supported in the Arduino IDE which are well provideded with libraries and example code.
The ADF5355 gets its very high frequency resolution by two fractional parts to programme its frequency. To have resolutions of the order of Hertz the micro controller must be capable of calculations with around 12 decimal digits which implies a 32 bit controller using double precision arithmetic is required. Details of the register programming are in the datasheet .
One suitable 32 bit Microcontroller is the WeMOS SAMD21 ARM Cortex M0 which runs at a 48 MHz clock rate. It is supported in the Arduino IDE and if programmed appropriately is capable of the necessary arithmetic precision. It is also available mounted on a pcb with a USB programming interface in the well established Arduino hardware format, Ref . Other options would be some of the Maple Leaf boards using ST Microelectronics STM32 chips.
For this project I choose a SSD1306 1.3 inch OLED display. These are available in either I2C or SPI versions. The 0.96 inch variant is more common but requires good eyesight! I sourced the 1.3 inch mounting PCB and the 1.3 inch display separately and assembled my own. Although the display is small it gives a very sharp image. Ref  and . I opted for the I2C bus for the display to keep it separate from the SPI bus controlling the ADF5355. I thought this wise to minimise noise transmission to the synthesiser. For a typical schematic of the of the OLED display module see .
To set the frequency and tuning step size Keyes type push button rotary encoders were used. These are widely available from the usual internet sources.
Two encoders were used, one as the main tuning control and the other to select the step size. The push button functions were used to select preset frequencies in each of the amateur VHF/microwave bands and to set the output power level.
The module I used is shown below. It used a 125 MHz reference oscillator with a push-pull output as that is claimed to improve the phase noise (PN) performance. For this application I used a external 100 MHz ovened reference for better frequency stability and accuracy. To provide a push-pull reference input to the board I supplied the reference via a small trifilar wound bulun transformer constructed using a double hole ferrite, connected to the REF+ and REF- inputs. To disable the onboard reference and connect the external one I had to unsolder L1 and populate R17 and R8 with 100pF capacitors. The schematic is available in the references at the end .
The ADF5355 gets its very high frequency resolution by using two fractional parts along with a programmable second modulus to set its frequency, (The first modulus is fixed in hardware). Frac1 is a 24 bit number and Frac2 is a 14 bit number. A good description of the process used to calculate the two fractional parts and the second modulus is given by Andy Talbot G4JNT in [ 2 ]. The Programme calculates these values from the frequency dialed into the OLED display and loads them into the appropriate registers. To have resolutions of the order of Hertz the micro controller must be capable of eleven decimal digits. The SAMD21 is a 32 bit device so when used with double precision arithmetic it is up to the job.
The code is available from Github as detailed below. To compile the code and upload it to a SAMD21 board start the IDE running and in the Tools menu select “Arduino/Genuino Zero (Native USB Port)” as the board. It may be necessary to install the software for this if it is not already there. To do this go to Tools > Board > Boards Manager and select “Arduino SAMD21 Boards (32 bits ARM Cortex-M0+)” to install support for the board. The signal generator code can be downloaded from here on Github, .
The finished unit
The outcome of this project is a signal generator with continuous coverage between 52 MHz and 13.6 GHz in 10 Hz steps. The step size can be changed in decades from 10 Hz to 1000 MHz. The display is an I2C OLED device which gives a bright well defined image. By adding amplifiers after the outputs from the chip the output power is boosted to useful levels for testing purposes. In terms of phase noise and spurious signal output it is not a match high end professional units but it sure beats then on cost, size and weight!
Addendum 1 - Using a Maple Mini Microcontroller instead of the SAMD21
Initial work on this project was done with the WeMOS SAMD21 board. This board is relatively expensive at around £8. A lower cost alternative is a Maple Mini board. These are around £3 and sport a 32-bit STmicroelectronics STM32F103CB ARM Cortex-M3 128k Flash, 20k RAM processor and are very adequate for this project. The pinning is different from the Atmel SAMD21 and there are slight differences in the code and the schematic. The code for the Maple Mini is available at  and the schematic showing the connections is below.
Recently two varieties of PCB have become available from different Chinese internet outlets which have AD ADF5355 PLL ICs on them. They are described as “54MHZ-13.6GHZ RF ADF5355 PLL Phase-locked Loop VCO Synthesizer Board”. One on 1.6 mm FR4 board with green solder resist and the other on a thinner board material which again appears to be FR4 but with black solder resist. Both have broadly similar circuitry on them which closely follows the design of the AD evaluation board for the chip, but there are some differences.
Figure 1. Green PCB
Figure 2. Black PCB
Both boards are available for prices starting at about £55 which is about the one off price for the ADF5535 chip in the UK. They can be made to work under the control of a PC via their SPI bus, [1,2], with the AD evaluation software. I have now had one each type of board to try out and they both functioned correctly but the phase noise was a bit disappointing and the output powers were lower than quoted in the data sheet. The lack of output power is easily overcome with a MMIC amplifier and hardly surprising considering the FR4 substrate, but the PN performance was considerably worse than suggested by the data sheet and merited further investigation. The schematic diagram of the black PCB can be downloaded from here,.
Phase noise reduction
My initial hunch was that the power lines were the source of the noise. There are two voltage regulators on the boards, one a LT176333 which powers both the 3.3 Volt analogue and digital Vdd lines and the other a LT17635 which powers the VCO and charge pump circuitry with 5V. This is powered directly from the output of the regulator and there is no further on-chip regulation, whereas the 3.3 V rails are further regulated down to 1.8 V on-chip. Both the LT parts are billed as “low-noise” in the data sheet, but I noted that the AD evaluation PCB design used AD parts which are considerably more expensive than the LT parts. Their respective data sheets reveal that the AD parts were about ten times less noisy. The simple expedient of adding a 3300 uF low ESR capacitor across C30 (on a black PCB) bypasses the noise from the regulator output quite effectively. I used a Rubycon MBZ series 6.3 V part with and ESR of 12 mOhm. The screenshots in figures 3 show the reduction in phase noise at 10 GHz. This also is effective on the green board.
Figure 3. PN before and after addition of Cap across C30 on black PCB.
In order to check whether the chips were now performing as well as they should I measured the PN and compared the result with a simulated PN plot produced by the free ADIsimPLL tool which is provided by AD. The results are shown below.
Figure 4. Measured PN at 10 GHz with the addition of extra decoupling
Figure 5 Simulated PN at 10.4 GHz with a 15 kHz loop bandwidth.
The data sheet does not give PN curves at 10 GHz but does show the behaviour at 5GHz . This suggests that the levels at 100 Hz, 1kHz and 10 kHz should be in the -75 to -80 dBc region which is borne out by the measurements. The conclusion from this is that these boards can be made to realise the full performance of the chips and can be useful RF sources for simple low power beacons, signal generators and local oscillators. They may also be useful for multiplying to the higher bands.
- “PC Control of Analog Devices ADF4XXX Synthesiser chips over USB”, https://gm8bjf.joomla.com/articles/9-pc-control-of-analog-devices-adf4xxx-synthesiser-chips
- ADF5355 evaluation software. http://www.analog.com/en/design-center/evaluation-hardware-and-software/evaluation-boards-kits/eval-adf5355.html#eb-overview
3. ADF5355 schematic for black PCB. http://gm8bjf.joomla.com/images/pdf/ADF5355_sch.pdf
4. ADF4351 schematic for black PCB. http://gm8bjf.joomla.com/images/pdf/ADF4351.pdf
Acknowledgement: Thanks to John Miles KE5FX for making his GPIB Toolkit freely available. This was used to make the various graphs in this article. See: http://www.ke5fx.com/gpib/readme.htm
Since writing the Scatterpoint article further investigation of the ADF5355 Board and an ADF4351 board has be carried out. This work has concentrated on the “black” PCBs labelled NWDZ. The schematic diagrams of these boards are here [3, 4]. For both boards it was found that adding extra low frequency decoupling the individual power rails gave further small improvements. For the ADF5355 board adding 1000 uF 6.3 V low ESR electrolytic across C29 (AVDD), C31 (DVDD) and C30 (VP5V) gave improved PN. At 12 GHz the PN is around -80dBc 100Hz off the carrier. For the ADF4351 board adding 1000 uF 6.3 V low ESR electrolytic across C30 (AVDD) and C31 (DVDD) gave much improved PN performance.
I have refined the PN measuring setup as I concluded that the noise out of the ADF5355 at 12 GHz was close to the PN floor of my HP8566B SA. I am now down converting the output (suitably padded off) by mixing it with the 12.5755 GHz output from a CTI-Herley PLL to 575.5 MHz and measuring the PN at the lower frequency. This means that the SA is operating on the fundamental of its YIG insead of its third harmonic. This drops the PN floor of the system by 20log3 = 9.5 dB which is a useful improvement. The output from tyhe CTI-Herley is much lower noise that the ADF5355. The plot below shows the result.
Blue trace - ADF5355 running with 25 MHz reference frequency doubled to 50 MHz PD frequency, output frequency of 12 GHz and Icp set to 0.3 mA
Pink trace - ADF5355 as above but with an Icp set to 4.8 mA.
Green trace - Elcom DFS1201 running at 11.600 GHz (for comparison)
Red trace - SA 100 MHz reference output to indicate the PN floor.
The close to carrier noise is improved by the higher bandwidth affored by the higher charge pump current value by 20 dB. All the measurements on the ADF5355 were made on a board which had all three power rails decoupled with 1000 uF 6.3 V low ESR capacitors as described in addendum 1 above.
If carefully decoupled and set up correctly these ADF5355 PCBs can give PN performance close to the now unobtainable Elcom DFS units and are useful as LOs for 24 GHz.
I decided to try using the ADM7150 voltage regulators recommended by AD. As mentioned above they have a noise performance almost ten times better than the LT parts used on the Chinese Evaluation boards. the ADM7150 is a fixed voltage regulator and comes in both 3.3V and 5.0V variants. Unfortunately the pinout is different from the LT1763XX parts on the PCB and they also use the dreaded pin 0 as a ground pad. To use them I decided it was necessary to design a small daughter card to fit over the "Black" board. The layout, schematic and the BoM of the board are below along with a couple of pictures showing the board in place on the original PCB.