DDS RF Signal Generator AD9914 3.5GSPS Shield for Arduino

We selected the AD9914 because it is currently the only commercially-available synthesizer capable of delivering a sine-wave output up to 1.75 GHz (when operating with a 3.5 GHz clock — first Nyquist zone). Furthermore, this synthesizer features extremely low phase noise — for example, the datasheet cites phase-noise around –128 dBc/Hz at 1 kHz offset from a 1.396 GHz carrier. Harmonic distortion (SFDR) is also excellent, with values better than –50 dBc.

In addition, the AD9914 supports programmable-modulus mode, allowing output frequencies that are not constrained to simple power-of-2 ratios of the system clock but instead arbitrary rational fractions.It also offers extremely fine frequency resolution (64-bit) — nominally down to ~190 pHz.

Analog Devices provides an evaluation platform (the “EVAL-AD9914 Evaluation Board”), which allows users to exercise the claimed performance of the AD9914 — but that board has some serious drawbacks. It requires four separate low-noise regulated power supplies to work correctly (for example, a pair of laboratory power supplies such as Keithley 2231A-30-3) and it does notinclude a clock source — you must provide an external reference clock. This makes the official evaluation board far from “plug-and-play”. In practice you also need a computer to control it, and you’ll need a fairly expensive setup including a low-noise oscillator and multiple power supplies.By contrast, while there are some low-cost Chinese boards that use the AD9914, they come without software, without detailed schematics, have silkscreen annotations in Chinese characters, and often lack documentation or vendor support — we found that such boards typically suffer from poor PCB layout, insufficient supply decoupling resulting in digital noise coupling into the analog path, and missing output-filtering which negates many of the advantages of the AD9914.

Our project goal was therefore to create a more accessible and functional board in the form of a shield for an Arduino Mega. Here are the key advantages of our “DDS9914” board compared to the reference board from Analog Devices:

• Works out of the box (no need for multiple lab supplies)
• Open-source firmware (hosted on GitHub)
• No additional wiring or ribbon cables required
• Does not require an external clock source

Because of these features the board is highly suitable for hobbyists and electronics enthusiasts, as well as for laboratories and researchers working in high-frequency RF systems.

To make the synthesizer work “out of the box” we mounted on the board five low-noise LDO regulators (one of which powers the TCXO) so that the board can run from a single 12 V supply with isolated supply lines. For the clock we fitted a TCXO which allows us to clock the DDS core up to ~2.5 GHz. This limitation is due to the internal PLL of the AD9914 — to unlock the full ~3.5 GHz capability one must supply an external reference at ~3.5 GHz (some AD9914 units run at 4.0 GHz). In our (v3.0+) version of DDS9914 we support software-selectable switching between internal TCXO and external reference via an on-board RF switch.

On the output we included a low-pass filter (LPF) to suppress the image harmonic and an output transformer to suppress even harmonics.

One could have simply purchased one of the Chinese AD9914 boards on eBay or AliExpress, but as mentioned many of them suffer from faulty design: poor routing, inadequate power decoupling (leading to digital noise coupling into the analog domain), missing output filtering, which in turn nullifies much of the benefit of the AD9914. We therefore advise against buying them, as they often yield disappointment and wasted time and money.

Here is a brief history of how we developed the DDS9914: it took us more than two years. We used the reference schematic of the AD9914 evaluation board as a starting point and placed it on a board sized for the Arduino Mega shield footprint (53 × 114 mm). Because of that we had to design a 4-layer PCB. On a relatively small board we managed to fit the AD9914, all five LDO regulators, an OLED display, a rotary encoder, an output-enable push-button, two ceramic LPF networks (for image suppression), the output transformer, two SMA connectors, and a cooling fan. We also included connectors for an external encoder and button in case you wish to mount the board in your own enclosure — eliminating the need to solder additional wires.

During development we encountered a number of interesting challenges. The first major issue was the heat dissipation of the linear regulators. We solved this by adding a step-down DC-DC converter ahead of the LDOs to reduce the voltage drop on them (thus reducing heat) and using larger-package LDOs (DPAK and even D2PAK) which further improved thermal distribution across the board.

Another critical choice was the output transformer — its bandwidth and matching determine harmonic suppression and flat frequency response. We selected a transformer with a lower limit of 100 kHz and an upper limit of 3.0 GHz (with ≤3 dB drop at the edges). Using such a transformer means you cannot go below ~100 kHz at the output — but if needed one can remove the transformer and replace it with a wire-jumper to bypass this limitation.

Internally, the AD9914 offers extensive clocking flexibility, so we also implemented convenient software support for these features: key settings are available via a graphical menu and controlled via a rotary encoder. During design we experimented with the Balun transformer 617DB-1022 to convert single-ended to differential clock signals, but it performed poorly with the AD9914 — instead we adopted the ECL clock/data buffer ADCLK925 which helps reduce harmonic content at the DDS output.

In closing: our firmware source code for the DDS9914 board is open source and actively developed (unlike the unsupported Chinese boards mentioned earlier). Beyond the core functionality, we are planning to implement support for the programmable-modulus mode (as mentioned above) in a future firmware update.