AD8306 Logarithmic Amplifier Noise

• DC output from the detector cells are generated with an averaging cap.
• We don't have data on VLOG Noise vs. Input Signal level on the AD8306, but in general for log amps of this sort, noise on VLOG decreases as you increase signal level to a certain point (different on each log amp), then does not change as you increase the input signal level.  An example of this can be seen in the ADL5506 datasheet on Figure 31.

Hi jdobler,

thank you for your reply.

• So the detector cell is really an "ordinary" fullwave rectifier?!
• Thank you for the hint of Figure 31 in the  ADL5506 datasheet. The decrease of noise contribution makes sense. In my opinion the decrease is caused by the limiting behaviour of the detector cells. So by increasing the input signal the single detector cells successively limit their output DC value (and also the noise), as the input RF signal is too high. Thus the "limited" cell doesn't add any further noise (additive noise for the cell remains constant). Is this consideration correct?
• Are there any values for the detector cells wigths available?

BR

Hi,

I'm currenty analyzing the the log-amp cascaded amplifier/limiter chain according to the AD8309 datasheet (page 7-9). The results are looking promising, however the resulting trend is shifted towards a lower output voltage. My logarithmic analyze leads to an output trend similar to the attached image (figure 22 on page - 9 in AD8309 Datasheet). My solid (curved) trend is about 0.23V lower than the dashed straight one.

The dashed trend follows V_out = V_Y * 20 * log10(V_in/V_X) with values stated in the AD8306 datasheet.

• V_Y = 20 mV/dB
• V_X = 10^(-108 dBV/20) =  3.98 µV

According to the AD8309 datasheet, the knee Voltage E_k, the voltage level where the amplifier/limiter changes its amplification from A (4 = 12.04 dB) to 1, is obtained by

E_k = V_Y * 20*log10(A)/(A-1) = 80.3 mV

So it looks like I'm missing some kind of offset, which is responsible for the shift of the trend. I can't figure aut where this additive factor comes from.

Has anyone an idea?

Concerning the 23 Aug. 2017 post:

The detector cell is a proprietary design, but can be thought of as rectifying the limited signal of the 12 dB limiting amplifiers.  The basic reason that the noise at VLOG goes down with input signal level is that as input signal level increase, the gain through the limiting amp chain decreases.  I think that is what you were trying to say in your second bullet point.

Concerning the 4 Sept. 2017 post: 

You are mostly observing the part-to-part variation of the intercept.  The specs table lists how much variation you can expect over part-to-part (16 dB) and temperature (18 dB).  You can easily calibrated this variation out with a simple 2-point calibration.

Hi jdobler,

thank you for the reply.

Concerning the 23 Aug. 2017 reply:

Thanks, yes that's what I tryied to say.

Concerning the 4 Sept. 2017 reply: 

Unfortunately this specs are only stated in the AD8309 datasheet, and not in the AD8306 specs. Nevertheless, by adjusting the intercept I achive the desired trend. To do so, the intercept was changed from -108dBV to -98dBV (one dB higher than stated in the AD8309 datasheet, for the temperature variation).

BR

You should expect similar intercept part-to-part variation on the AD8306 as is listed in the AD8309 datasheet.

Thank you, I already thought so!

Hi jdobler,

I made some measurements of a realized protoype, to compare my analytic calculations with measurment data. The noise acts as expected, by means of a decreasing noise for higher RF signal levels.

However, at higher input levels the measurement results differ from the analytic one, by means of a higher constant noise level. I have not considered the additional four detectors direct at the IC input (4 x DET). Those detectors make use of a passive attenuator (LADR ATTEN), which may add additional noise.

Would it be possible to provide some information about the first four detector cells (4 x DET) and the attenuatior (LADR ATTEN), to add them to my analytic model?

BR