AD9850 and AD9851 Output Calculations

These are done using the method shown for AD9951, using simple logic.

This analysis was prepared in response to an emailed request from Robert G3WKU:

> To return to a very old theme of mine........I still don't think we have
> any satisfactory explanation as to why it was that the original
> Pic-n-mix with the DDS 9850 actually did produce enough drive.  Stars
> were ok until they changed to the 9951.

The following observations and sums resulted...

AD9850 and AD9851

The datasheets for these devices are quite similar, but the AD9850 has a maximum current in the output settable to 10.24mA with 3k9 or 20.48mA with 1k95 versus the AD9951 with 10mA using 3k9 and a maximum of  15mA. Output compliance is specified as 1.5v for AD9850 compared with AVdd+-0.5V for AD9951.
The apparently positive-only output compliance for AD9950 suggests that transformer coupling may not be easy, since it would tend to position the output symmetrically around ground.
AD9950, the original DDS of the 'Pic 'N Mix' Synthesizer
On the original Pic-N-Mix, only one output from the AD9850 DDS is used, referred to ground by a 200R, since the 9850 sources current (conventional flow + to -). This is the same as the evaluation circuit shown in the datasheet.
The output maximum current is set to 10.24mA using 3k9 on pin Rset, so the output is a 0 to +10.24mA current waveform. The 200R resistor connected between output and ground provides a d.c. path, without which the device could not work. The 10.24mApp output current is converted to a voltage waveform by the 200R resistive load in parallel with the 200R filter impedance, to feed the 200R filter and Buffer Amplifier at 1.024Vpp.

The STAR and Pic 'N Mix reconstruction filter has always been exactly the design from the 9850 application board, except that the inductor on the input side is 910nH on the app. bd and 1uH on the Star.
AD9851 and enhanced Pic 'N Mix
Analog Devices introduced a new DDS part, AD9851. This is pin-compatible with AD9850 and has a few additional features. There is an internal optional digital 'x6' PLL frequency multiplier, together with increased operating frequency, now up to 180MHz at 5V supply rather than 125MHz. It was used as a drop-in replacement for the original AD9850, and permitted the use of a higher reference clock with correspondingly better output level at high output frequncies (sin(x)/x reduces it by less). 

The original Pic 'N Mix design has no output transformer at all. In the quest for greater performance, a modification to add a centre-tapped transformer was intoduced. This uses both DDS current outputs and provides cancellation of even harmonic distortion. This used a large Epcos binocular core, with 3+3t:12t, "mounted in lieu of the 100R & 200R on the DDS Carrier" and is heralded as "this gives 6dB more LO output".

This transformer is later changed in the Official Mods to 3+3:8t. Also, in this doc, it is stated "A BN43-2402 core is at least as good as the [Epcos] 62152 core recommended", since many people were having difficulty obtaining the Epcos ferrite core. A 56R terminator on MR was added in this mod-set also, previously the H-mode squarer had run unterminated (and hence at twice the input voltage level).

With the 'transformer mod', there is no source-termination at the DDS. The transformer provides the d.c. return path for the DDS output currents and provides an output symmetrical around ground.  The signal termination is the 200R load terminator at the Buffer Amp input, reflected back through all the interim (filter, transformer) onto the DDS output pins. This would give more output, as stated in the modification notes.
The stated compliance range for 9851 is -0.5 to +1.5V; probably the -0.5 is limited by substrate diodes conducting so would probably not damage the chip.

The AD9851 data-sheet says of the outputs "The true output of the balanced DAC. Current is sourcing and requires current-to-voltage conversion, usually a resistor or transformer referenced to GND. IOUT = (full-scale output–IOUTB)."  In contrast, the AD9851 data-sheet makes no mention of transformer output and appears unable to use it (the output compliance would need to extend negative of ground but is not specified to do so).

The AD9851 data-sheet also states "The differential DAC output connection in Figure 9 enables reduction of common-mode signals and allows highly reactive filters to be driven without a filter input termination resistor (see above single-ended example, Figure 8). A 6 dB power advantage is obtained at the filter output as compared with the single-ended example, since the filter need not be doubly terminated."

Fig 9 from AD9851 datasheet

In other words, there is no need for source termination in this case; the only terminator is the reflected load.

It can be seen that the AD9851 transformer design did not use source termination if it had a transformer, giving rise to the "6dB power advantage".
In summary:
* The Original AD9850 design has no transformer; 10mA source into 200Rload //200 filter gave 0-1V output, 1Vpp into the filter, no transformer losses. The 200R source load is necessary for d.c. conditions.

* AD9851 with transformer had initially 3+3:12t, later 3+3:8t, but with no resistive loads on the DDS pins; the reflected ultimate load is the only load.

* AD9851 has greater overall compliance range +1.5/-0.5. This means the original total load of 100R to ground kept well within compliance range (0 to +1v from this circuit), tolerating impedance ripple in the filter.

* Taking, for now, only the 3+3:8t transformer eventually used with AD9851, there is no source load (no resistor, just the Hi-Z DDS output) so each output can be said to see the Zload according to its own turns ratio; there is no primary terminator to share by transformer action. The 200R final load reflects to the outputs as 28R per output (3^2 / 8^2) * 200R. Because there is no source terminating resistor, there is no effective paralleling of resistances as there is in the AD9951 design; each half-primary sees the other half in inverse parallel, due to transformer action, but each presents an 'infinite' load since it is a current source. Thus 10mA drive gives 280mV on an output at each +ve peak; this reverses polarity due to the transformer inversion when the other o/p drives 10mA and gives 560mVpp at each output, or twice this between them; this transforms up by 8:3 to become 1.5Vpp into the filter load of 200R.
This is 50% more drive than the 9951 circuit.
It is well within the AD9851 -0.5V compliance limit as long as the load is present.
If the load is absent, the current has 'nowhere to go' and the device will conduct it to wherever it can (probably to parasitic substrate diodes on the -ve swing).

* Using the same 'simple' calculation as above; with the original AD9851 transformer of 3+3:12t, the corresponding per-output DDS load reflected from the 200R filter would be 12.5R, so 10mApp gives 125mVpp, the transformer primary sees +-125mV per half-primary, which is transformed up by the transformer ratio of 12:3 to +-500mV or 1Vpp into 200R filter. So the transformer ratio change from 3+3:12 to 3+3:8 actually produces 50% more load voltage! The output pin swing is small, not making use of the compliance range.

Bottom line

If my sums in the -um- sum-mary above are correct, the AD9851 (and maybe AD9850 if the compliance range actually extends below ground, although this is not in the datasheet) with the corresponding most recent recommended transformer mod (3+3:8t) drives 50% more voltage into the filter than does the present AD9951 circuit. Anyone still using the single-ended AD9850 output will have nominally the same filter input as the AD9951 design, but without the transformer losses of the AD9951 design.

Removing the 100R output loads from the AD9951 transformer primary would greatly increase (double) the output, but would be on the verge of exceeding the compliance range and would definitely do so if unloaded, or if the load were higher than 200R at any frequency. The input impedance of the filter does in fact rise considerably (to 300R) at high output frequncies... The AD9951 is a low-voltage tiny geometry part and will not take over-voltage without damage. Peter G3XJP has said that AD9951 devices were destroyed during development; this is quite probably the reason.

We can state, then, that the lower LO output for AD9951 users (1Vpp vs 1.5Vpp) is due to the source-matching used with AD9951 that is absent on AD9851 with the transformer mod. Power is lost in these resistors, for the same DDS current swings.
The AD9951 source-terminarion is essential to avoid damaging the device.


Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.