AD9851 DDS ASK Modulation linearity test

It was suspected that the AD9851 DDS (Direct Digital Synthesizer) output voltage was not linear varying with the input DC control voltage, if so, that may contribute to the problem of the pulse distortion that was previously discussing, a test is done to verify this.

DC voltage is applied to the current reference input. The RSet pin is holding a 1.26V voltage, so when applying a 1.26V in DC input point, there would be no current flow in/out, that gives zero output. when DC input is 0V it's the maximum output, which is around 0dBm.

34 data points is recorded and the data is plot in DC control voltage VS RMS voltage of RF output of DDS. The conclusion is, well, the relationship IS linear. 

When bypassing the power amplifier stage, so that the DDS is feeding to the antenna directly, that gives the receiver a nice waveform, with almost no overshoot. 

The overshoot problem is narrowed down to the power amplifier stage.

Overshoot issue investigation

The overshoot issue is improved by reducing output power. This add one more evident that the nonlinearity of power amplifier stage contribute mostly to the rise/fall time asymmetry issue.

Left : Receiving waveform from full power transmitter

    Right : Receiving waveform from ~half power transmitter (overshoot is reduced)

The Bessel filter could not yet solve overshoot issue

The Bessel pulse shaping filter is patched to the 27MHz, AD9851 DDS remote control transmitter.

The pulse shape looks OK when after shaping filter, but the modulating waveform envelop rise time not matched with shaping filter output. The rise time of envelop is a lot faster than fall time, it contribute to the overshoot in receiving waveform.

The ASK demodulator in receiver also have an old issue of envelop rise time faster than fall time. This makes the issue getting even more serious.

Demodulated waveform in receiver, sadly it has ~10% overshoot, it is highly suspected that this is caused by nonlinear phase response in 455kHz ceramic IF filter, the issue still not yet improved by adding transmit pulse shaping filter. The problem looks like from the AD9851 current reference input having asymmetry in rise time and fall time. Or power amplifier stage amplitude nonlinearity. This overshoot issue must be resolved.

Search for ways to reduce pulse distortion

The goal is SAY NO TO FIRST ORDER RC FILTER , may be start with analog filter approach. The Bessel filter and Gaussian filter both have good phase characteristic, that should improve the time domain response, make them suitable for pulse shaping application. There's are good filter design tools out there, Analog devices/TI both provide web based tools, but TI required account login. 

The Analog Devices ANALOG FILTER WIZARD hided some filter terminology such as Butterworth/Chebyshev/Bessel filter type, in another words, they use Fewest stage/Fastest settling to abstract out the optimization goal. Low pass filter with 5kHz cutoff frequency specification is picked.

It's a very relax frequency response specification, so it only takes 1 stage of opamp, it also give option for optimize for noise or optimize for power. Low noise is chosen in this test. Sallen key topology is default.

Simulation in LTSpice, using LT1797 opamp.

The filter was firstly implemented using LM358 general purpose opamp, but it give weird spike in the rising of the pulse, then it is replaced by LMH6646, which give a nice waveform, the reason why LM358 is not working is not yet understood. The waveform match with the simulation. The waveform is more symmetric than a first order RC filter, nice.

The ugly face of first order RC filter

In digital radio transmitter, the baseband waveform needed to be filtered, need to be lowpass filtered, because the bandwidth of a rectangular pulse is infinite and modulating an carrier with an infinite bandwidth baseband waveform producing infinite bandwidth signal, that cause interference to near channels. 

The first order RC filter, which is the simplest lowpass filter, and we using it just everywhere. It has cut-off frequency at fc = 1/(2piRC), but it roll-off at -20dB per decade which is very gentle, so it actually not "cut-off" after fc, it just slowly taking out the high frequency component inside a signal.

But now what make me even more frustrated, is the response of first order RC filter, in time domain.

First order RC Filter ! Why you are NONLINEAR PHASE ?

The pulse became asymmetric, it jump rapidly in beginning and settle slowly at the end, and the long tail extended to next symbol, that means some energy is leaked to the next symbol, sometime we call it Inter Symbol Interference - A bad guy. The problem of asymmetric waveform is caused by unequal delay of different frequency component inside the rectangular pulse, with lower frequency delay more and high frequency exit faster, textbook call it nonlinear phase, I call it ugly.

It's wont work, something have to be done.

Analog Devices AD605 X-AMP test

Introduction from Analog Devices AD605 Datasheet:

"The AD605 is a low noise, accurate, dual-channel, linear-in-dB
variable gain amplifier (VGA), optimized for any application
requiring high performance, wide bandwidth variable gain
control. Operating from a single 5 V supply, the AD605 provides
differential inputs and unipolar gain control for ease of use.
Added flexibility is achieved with a user-determined gain range
and an external reference input that provide user-determined
gain scaling (dB/V)."

It sound's very powerful, if this can replace the discrete transistor version of VGA in my receiver projects, it save a lots of board spaces and greatly improve the dynamic range. Let's make a evaluation board to try try.

Setup:

The AD605 evaluation board, a 68 ohm resistor is loaded with the input to make up of a near 50 ohm input impedance.

Test setup: Gain control voltage is from adjustable power supply. From 0.1V to 3.0V.

Signal source is from signal generator, 455KHz, -60dBm, quite a small signal.

Output power is measured from spectrum analyzer, it's dynamic range way better than an oscilloscope.

Gain-Control voltage plot from AD605 datasheet. In this test, FBK(open) configuration is using.

Result:

Gain-Control voltage plot from actual test, which is very closed to the plot in datasheet. A nearly linear-in-db curve between 0.1V to 2.5V control voltage. and tuning range of about 45dB. Certainly a very respectable figure.

45dB of gain is not enough for a typical receiver, there are another amplifier unit in the AD605 package, may be I could cascade them and put a ceramic filter in between them, and share same control voltage.