How to improve PA shoulder

[sedighe]

Hi friends

I designed a PA to be used with OFDM signal, but the shoulder is about 10dB worse than what I expected regarding to transistor's datasheet, I replaced the aluminum pallet with copper one but no improvement was seen! can anyone guide me how else i can work on my PA shoulder?

Regards
Sedighe

 

[goldsmith]

Hello
What is your frequency , and which arrangement did you use for amplification?
Best Wishes
Goldsmith

 

[RF_Jim]

Hi friends

I designed a PA to be used with OFDM signal, but the shoulder is about 10dB worse than what I expected regarding to transistor's datasheet, I replaced the aluminum pallet with copper one but no improvement was seen! can anyone guide me how else i can work on my PA shoulder?

Regards
Sedighe

sedighe,

Late in the design of the HPRF (High Power RF, the RF PA deck) board for our P4 WiMAX product, we discovered insufficient capacitance for low frequency bypass on the Vcc (Drain voltage or Bias supply) which required re-work to add capacitance ... after adding additional Vcc bypass capacitance the 'spectral mask' performance improved, including EVM. It is critical to provide adequate capacitance to the drain bias supply as the 'peak' current draw at various points in time can be quite high with a multi-carrier OFDM signal.

Please let us know the result if this works to improve the situation.

BTW, adding capacitance affected the series 'switch' that was used to switch the DC to each of the HPRF boards, and necessitated the incorporation of "slow-start" circuitry to ease the rush of current when the fairly high-valued Vcc bypass caps charged during HPRF turn on otherwise our power supply output voltage would 'dip' and present other system-level problems (like uC reset).

Jim

 

[sedighe]

This is a wideband PA in 400~900MHz, and I used Balanced topology with cable BALUN.

---------- Post added at 07:43 ---------- Previous post was at 07:40 ----------

sedighe,

Late in the design of the HPRF (High Power RF, the RF PA deck) board for our P4 WiMAX product, we discovered insufficient capacitance for low frequency bypass on the Vcc (Drain voltage or Bias supply) which required re-work to add capacitance ... after adding additional Vcc bypass capacitance the 'spectral mask' performance improved, including EVM. It is critical to provide adequate capacitance to the drain bias supply as the 'peak' current draw at various points in time can be quite high with a multi-carrier OFDM signal.

Please let us know the result if this works to improve the situation.

BTW, adding capacitance affected the series 'switch' that was used to switch the DC to each of the HPRF boards, and necessitated the incorporation of "slow-start" circuitry to ease the rush of current when the fairly high-valued Vcc bypass caps charged during HPRF turn on otherwise our power supply output voltage would 'dip' and present other system-level problems (like uC reset).

Jim

I checked the datasheet again and I saw that they recommend 4.7uF in DC line, I didn't imagine that may affect the shoulder, I am going to check it out today and will inform you dear Jim

 

[goldsmith]

Again HI
What did you use as PA ? here is many simple ways , that you can do to obtain good specifications. is your PA linear ? or non linear?
Best things
Goldsmith

 

[biff44]

DC bypass capacitance has to be sized based on the nature of your modulation. If your modulaiton is non-constant envelope and broadband, say 100 MHz wide, your external capacitance can be more modest. But if your modulation is narrowband, like 100 KHz wide, then the time domain envelope will be varying at a 100 KHz rate, and a small bypass cap will be insufficient. Depending on your specific requirements, you might need 1000 uF or more!

Of course, other things can cause high intermodulation products.

How you are driving the amplifier may effect things. If you are overdriving the input with too much power, then the 3rd order products will rise. Drop input power by 1 dB should improve the IP3 products by 2 dB, in general.

You might have a mismatch, especially at the output, that is screwing amplifier, either at the fundamental frequency or at its harmonics and therefore changing the IP3 point. Try varying the output components or their lenght from the amp (especially for lowpass/bandpass filters) and see if it has an effect.

Many amps have many stages. They all add to the intermodulaiton products. Your cascade design should be done so that the very last stage contributes the most to the intermods, and the input stages contribute very little.

Rich
Maguffin Microwave LLC

 

[vfone]

Bias decoupling at modulation frequency, and output impedance optimization, both can affect the linearity of the PA (shoulders or ACPR whatever you name it).

But choosing the right DC bias point of the amplifier is the most important parameter that would affect the linearity, and this is the first thing that you have to check.

 

[RF_Jim]

Bias decoupling at modulation frequency, and output impedance optimization, both can affect the linearity of the PA (shoulders or ACPR whatever you name it).

But choosing the right DC bias point of the amplifier is the most important parameter that would affect the linearity, and this is the first thing that you have to check.

It might also be a good thing that one first adhere to the app note recommendation of proper DC bypass (as sedighe noted above); any amount of 'DC bias point tuning' will not correct for a missing, necessary, critical-to-high-performance operation component!

Next in line, as some have pointed out, would be:

1. DC bias point (as one watches Drain current across Drain current sample resistor as the Gate Bias voltage is adjusted) followed by

2. 'Tuning' the drain matching network; in production, we have found that a fixed-value design components can 'wander' from design center, with early production run units meeting spec, and as time proceeds 'things change' and one begins see fails for ACPR on say, the high frequency end of the band (we tested at a Low, Middle and High frequency in a particular band). As it turned out out, the FET manufacturer admitted to 'losing the magic' to make repeatable FETs but that did solve our issue, so we had to by some trial and error add add or subtract small amounts of capacitance to meet ACPR spec across the band. A .5 or .7 pF cap was necessary to bring the drain output matching network back into the 'tuned' state. This was using micro-strip of various widths (around 1 cm) at a frequency of 3.5 GHz and power level (for a multi-carrier OFDM PA) of 30 dBm RMS power (with much higher pk power of course).


Jim

 

[RF_Jim]

This is a wideband PA in 400~900MHz, and I used Balanced topology with cable BALUN.

---------- Post added at 07:43 ---------- Previous post was at 07:40 ----------




I checked the datasheet again and I saw that they recommend 4.7uF in DC line, I didn't imagine that may affect the shoulder, I am going to check it out today and will inform you dear Jim

I should add: Even in this day and age of multi-carrier capable ARB signal generators, sometimes there is no substitute for a two-tone test; we were able to detect a PA stage instability on a new design introduced to the contract manufacturer using two-tone test and the spurious output was obvious using the two tones whereas with the 'rectangular' multi-carrier spectrum signal simply showed an increased ACPR (an above spec value).

So, also assure your design is stable 1) with power level variation (two-tones kinda assure this and it's easy to see instability) and also 2) across the intended frequency band.

We saw this instability on this new design only at the upper end of the frequency band and when driven with RF power. Sitting quiescent (no RF drive) the circuit was stable (nominally Class AB1 bias). Since this was the last stage of about 4 stages in a TX/RX module with the circuity including input switch, an output circulator etc it was impossible at the time to simply do S11 S12 etc msmts to observe feedback margin in that last PA stage without 'cutting' up any of our customers production boards.

Jim

 

[sedighe]

Hi dear friends

I added two 4.7uF in each VDD line and shoulder in most of the band get better about 10dB , but there is still shoulder problem at the 60MHz end of the band (800MHz~860MHz) that I think maybe it can be improved by working on output matching. previously I had 20dB shoulder in the whole band and now I have about 30dBc and 26dBc for end of the band.

My PA is working in AB class, and I had checked its stability before and I solved it by adding a gate resistor, now the PA is completely stable. regarding to what "Rich" mentioned above I should say that the gain is normal (linear gain) which shows that I do not overdrive the device.the result of 2-tone test was also acceptable.

Thanks a lot for your helpful advices
Sedighe

 

[goldsmith]

Hello my friend .
Why class AB ? at this range of frequency , the class C or D or E or F are more better than Class AB . isn't it ? it is very simple to make high powers with best efficiency and low THD and and earn good specifications and cheaper than class AB !
If you want i can guide you with these classifications.
Best Regards
Goldsmith.

 

[xylon89del]

Hi all,
just curious, what is actually meant by "shoulders", which you all use above?

 

[RF_Jim]

Hi all,
just curious, what is actually meant by "shoulders", which you all use above?

When looking at the spectrum of a multi-carrier OFDM signal, the many carriers that make up the signal on the screen make the spectrum on the screen look "square", like a (TV cartoon character) Bart Simpson head, and the area where the signal should fall to 'zero' (near baseline of noise floor of spectrum analyzer) either side of the "head" that area is called the 'shoulder'.

An amplifier exhibiting IMD (Inter-moduation Distortion) will show a signal level above the baseline or noise floor on the spectrum analyzer, possibly into an adjacent channel, and why sometimes this is known as an ACPR (Adjacent Channel Power Ratio) measurement as well.

Regards,

Jim