Evaluating spurs and harmonics generated by DDS chips

I have seen much misinformation published and emailed to me regarding the performance of Analog Devices DDS chips. And seemingly learned hams have accused the manufacturer of these chips of falsifying data, or outright incompetence.

If I was to test any manufacturers chips, and was unable to achieve the same performance they publish, my first reaction would be "what am I doing wrong?" I would not declare them useless as one arrogant Ham/Engineer from Michigan did to me in an email.

I can't comment very much on the other chips, but I will talk about the AD5932 and AD9834 which are excellent chips for the price. As far as I know, none of these chips will drive a 50 ohm load directly. Some make the mistake of testing them with a 50ohm load, as I did the first time I tested them. As it turns out that makes huge difference, which isn't too surprising. After all connecting a 50 ohm load to a part designed for a 200 ohm load is obviously going to draw much more current from the output driver and likely cause it to operate in a non linear manner, thus producing harmonics and other forms of distortion. In order to connect my DDS to 50ohm test equipment, I made a simple resistor network that provided a 200 ohm load to my DDS, and presented a 50 source to the test equipment.

The second area is error is excluding the reconstruction filter in the test. The output of these DDS chips is the output of a D/A converter. But D/A converters are incomplete if they are not followed by a reconstruction (not anti-alias filter) filter. But for some reason people run tests without the filter, and not to surprisingly get bad data. My DDS board has a 5th order Chebyshev reconstruction filter. This filter is far from perfect, but it allows operation up to 40% of the clock frequency with good performance. DDS chips have a predictable mirror image signal. The purpose of the reconstruction filter is to attenuate the mirror signal. For example if you are using my DDS with an 80MHz clock, I recommend use up to 30MHz. With a 30MHz output and a 80MHz clock there will also be a mirror signal at 50MHz. With the filter I supply this mirror signal is down 50db. Now as you lower the output frequency, the mirror signal increases in frequency, making it easier to attenuate. So at 28MHz output, the mirror is at 53MHz, and down by 58db. At 25Mhz, the mirror, now at 60MHz, is down more than 75db. The 50MHz version of my kit is not quite this good. At the specified maximum frequency, 20MHz, the unwanted spur (at 30MHz) is only down 40dB. You have to drop down to 18.5MHz to get 50db spur attenuation.

Another problem area is the digital output provided on most of these chips. This is a high level square wave output that is at the same frequency of the output signal. Since it is a square wave, it is full of harmonics. If one is not careful picking off the test signal, then harmonics from the digital output can get coupled in. With the AD5932 there is an internal control bit that allows you to turn this output off but in my design it is on all the time, but only goes a few mm to a test point. Frequently I pick up the DDS signal by simply laying a wire loop connected to my radios antenna input near the board, and just listen for the signal on my regular radio. But doing this you are also going to pick up harmonics from the digital output. I did make a temporary change to the firmware to turn the digital output off. And when picking up the signal with a pickup wire, this made a huge difference in harmonic content. But when I ran a shielded cable from the output connector to my test equipment, I found that turning the digital output on or off made no difference.

PCB layout. It is very important to develop a PCB with good grounding techniques and proper bypass capacitors. You will find 8 bypass capacitors on my board, connected as the manufacturer recommends ,and then some. Included are special ones for the internal 2.5v supply in the DDS chip, and the D/A bias. I probably have 2 to 3 times as many caps as I really need, but when I did this sort of thing for a living, I was known for saying that I never saw something malfunction because it had too many bypass capacitors, but I've seen plenty of weird problems on boards that didn't have enough. Separate digital and analog ground planes are also necessary.

Doing all of this, my tests show the same or better results than Analog Devices published data. If you want the details please refer to analog devices data sheet. Manufacturers Data: AD5932.pdf AD9834.pdf In some cases, specifically with thier tests at 4.800MHz and 7.143MHz, my results were 10-20 db better than what they show, I am confident that the performance of my DDS will meet or exceed the worse case data published in the manufactures data sheet, in normal amateur radio applications.

The graph on my web site is one I picked out that's great looking for marketing purposes. But there are many others in the data sheet that aren't as great looking. But keep in mind that this DDS is a low cost kit, it is not intended as the best possible DDS in the world. It's great for many applications, but if you are trying to design a radio that has better performance than the Ten-Tec Omni I suggest you look for a better DDS, and expect to pay much more for it.