CC2590: Transmission output impedance?

Hi Albin,

I've notified a HW expert to look into this, they'll be returning soon (around mid next week), please stay tuned.

Thanks,
Tarek

Ilkka,

I can confirm the DN037 application note is the reference that you must use for the CC2590. The lower voltage will indeed affect the performance of the amplifier, most dramatically on the current consumption. However, given there are no characterization curves at 2.3Vdd, I don't have accurate numbers for this operation point.  

Hope this helps,

Rafael

So, it appears that the output impedance of the CC2590 is effectively undefined, or at least not specified with sufficient clarity to support matching optimization for non-reference designs. This makes the process of designing and tuning a proper matching network for custom boards almost impossible.

Based on experimental measurements, I suspect the CC2590 includes some form of feedback mechanism at its RF output — possibly envelope detection or directional coupler circuitry — which dynamically adjusts the output drive. It seems to aim for a consistent output power or gain, compensating for mismatches by increasing current draw when reflections occur. While I can't confirm the internal implementation, this would explain the behavior I observed: the PA draws significantly more current when the antenna is poorly matched, and less when matching is optimized. (In fact, I got some sense with the matching with very fast and accurate supply current measurements over the frequency spectrum. When tuned properly, the drawn current from the supply drops significantly.)

This strategy may help maintain functionality across a wide range of conditions and antenna designs, but it also means the output impedance can shift dynamically, making external matching less predictable. For battery-powered devices, where every millijoule counts, this can lead to notable inefficiency and performance degradation.

Hi,

What transceiver/SoC device are you using with the CC2590? 

Looking at TI Design Note DN037 (swra375), particularly Figure 4.1 (right side), it seems the maximum output power is achieved at an impedance noticeably more resistive than 50 Ω, assuming the chart is normalized to 50 Ω. This doesn’t align well with the 33 – j7 Ω assumption — unless the chart is normalized differently, which is unclear from the figure. Can anyone confirm what the Smith chart in DN037 is actually referenced to?

DN037 specifies that the measurements in Fig. 4.1 were made at the SMA connector reference plane, i.e. after the impedance matching/filter network. The (33 +/- j7) Ohms discussed in other threads (e.g. https://e2e.ti.com/support/wireless-connectivity/bluetooth-group/bluetooth---internal/f/bluetooth---internal-forum/128791/cc2591-customer-has-requested-s11-data-for-impedance-matching-circuit-design-between-antenna-and-pa-lna-zephyr-md-usa/477733?tisearch=e2e-sitesearch&keymatch=%2525252520user%252525253A4458#477733 ) is referring to the target impedance at the device pin reference plane. You can assume that Smith chart plot is normalised to 50 Ohms in this case.

Are you using your own custom design when experimenting? 

Regards,

Zack

We are using the CC2520 radio on a custom board with our own antenna design. Communication works reliably, but we are peak current–restricted (absolute max. ~50 mA for the full board). As a result, we've already limited the CC2520 output (currently using TXPOWER = 0x81).

We would like to optimize RF performance by better matching the antenna to what the CC2590 prefers to see at its output. If this impedance was known well, it should be a pretty straight forward job to do the matching. Currently, I have no confidence that it is e.g. the suggested 33±j7 Ω.

From your reply, I now understand that the dark red region in DN037, Fig. 4.1 represents favorable antenna impedances when connected to your reference board, and that the plot includes the effect of the built-in matching/filter network + parasitics— i.e., it does not directly describe the CC2590’s output impedance at the device pin reference plane.

While trying to tune towards the 33 ± j7 Ω target (from earlier threads), I found that the output spectrum from the DUT appeared relatively flat across the Zigbee band. That initially puzzled me — but it now seems likely that the CC2590 includes internal regulation or feedback, targeting constant gain regardless of mismatch. The tradeoff appears to be increased or decreased supply current draw, depending on the load impedance. I have not seen any TI documentation about this 'feature' though.

This interpretation is also supported by our measurements: when I happened to hit a favorable impedance (resonance) point, the current drawn by the PA dropped significantly. So going forward, I’ll continue tuning with both spectrum/network analyzer feedback and precise supply current monitoring. Lots of instrumentation is needed, but that seems to give the most direct clue about what the CC2590 “likes to see.”

BR,
Ilkka

Hi Ilkka,

The transceiver/SoC used will change the input matching to the CC2590, which will then change the required CC2590 output matching requirements (along with the supply voltage used).

There isn't any internal feedback mechanism; you can have the same Output Power with increased/decreased current consumption based on different loads presented to the PA output. If plotted on a Smith chart, these will trace contours of constant Output Power (with varying current consumption) - very similar to what you see in DN037 (but at the CC2590 output device pins).

The optimum target impedance will also change over the frequency band, but unfortunately I don't have the load-pull data for that available - which is the crux of the issue here.

Have you simulated the RF path of the TI reference design (i.e. including both component and PCB parasitics) and compared the load impedance presented to the CC2590 (over the target frequency band) compared to your design? There would be differences because of the different supply voltage, but that would provide an indication for how to approach the tuning process. The target load impedance trajectory shouldn't differ that significantly.

Regards,

Zack

Hi,

Thanks, this was informative. CC2520-CC2590 connection is straight and short with one DNP shunt reservation between the differential lines. I figured from other posts that straight would be ok, although T-filters would be better when tuned to the layout. I didn't realise that the transceiver side would affect the PA output impedance... There's not much that can be done with the one (un)populated shunt?

So, no active internal feedback, but simply better power transfer efficiency in certain areas.

I already had backwards simulated the reference design matching circuitries (2 of them actually), only the components without guessed parasitics to get a rough idea, but now I see that I might have had a wrong basis from the misinterpreted DN037, Fig. 4.1.

I'm seriously running out of time with this, but I'll probably do another simulation round from the reference design(s). I know the parasitics may have very big effect, so I'm not sure whether it will actually help so much, but I'll give it a try.

BR,
Ilkka

If time is a major constraint here, just focus on the output network. The different supply voltage is a larger design factor.

One point to highlight is that the general behaviour of the load-pull plots in DN037 will still be seen. Even if the locations of the Output Power and current consumption contours change, the concentric behaviour of the Output Power contours and the trend of low-to-high current consumption regions will remain similar to the plot in DN037:

Each frequency measured will have slightly different optimum regions; I'd suggest focusing only on the low, mid, and high frequencies of the band given the time pressure.

You can roughly "map" the contours using your idealised simulations of the output network. If you reach a region of unacceptable performance in one impedance trajectory (e.g. by increasing/decreasing a particular component value), note the impedance trajectory measured on the Smith chart using the basic simulated networks. That would build a (very) rough picture of the performance across the band. It's not perfect, but you can then at least try and avoid targeting the same load impedance repeatedly with different BOM values.

Another design concern here is stability, so please bear that in mind when tuning the output network (if you can check the spectrum). This would be a major design consideration even with all of the required load-pull data over frequency, voltage, temperature, etc. being available.

Regards,

Zack