Direct RF radio

Started by jekain314 1 month ago13 replieslatest reply 1 month ago215 views
I am looking at the pre-release Analog Devices AD9084  -- APOLLO MxFE. This has 4TX and 4Rx and ADC/DAC at 20/28 Gsps. There is also a less capable but similar in-production chip AD9082. The AD9084 has on-chip DSP and claims it can do a 192-tap FIR filter at the sampling rate. Can I use the FIR with this chip to replace the bandpass filter that normally occurs after the antenna and LNA for a receiver at, say, 2.4GHz? The radio would be the broadband antenna plus an LNA and the AD9084 sampling the RF signal at 20GHz. The FIR on the AD9084 would do the bandpass filter to restrict the signal to 2.4GHz for further processing. The same radio could also do GPS by switching the bandpass filter to the 1.5GHz for GPS. Would this DirectRF solution make sense?


[ - ]
Reply by mfuerterJune 13, 2024


Assuming I understand your question: You will still need to protect the AD9084/AD9082 from overload.  The purpose of the [hard] RF bandpass filter is to provide such protection.  Since the FIR is implemented in the Digital Domain, the A/D process is still wide open to any strong out-of-band incoming signals.  Such strong signals can cause crash [overload] of the ADC.  The FIR will act as a sub-band filter to reduce the thermal noise bandwidth and provide sub-band channelization [but this is downstream of the A/D conversion process].

Hence, you will still need the RF Bandpass Filter [preselector], even with the FIR.  The RF filter will be prior to [upstream of] the LNA.


[ - ]
Reply by jekain314June 14, 2024

My question came from considering the goals of SDR to move the ADC closer to the antenna (e.g., Direct RF sampling) -- also the youtube article below that discussed the 40 frequencies that must be treated by the iPhone for global access (each band has a separate custom FBAR filter). If the multi-application radio has to have a bandpass filter (hardware component) then we can not achieve the goal of a fully digital radio that can do, for example, the 802.11ax (2.4, 5, 6 GHz) and GPS (1.575, 1.227, 1.176 GHz) using just a wideband antenna and an LNA followed by the ADC. My thinking was the ADC (AD9084) can do 12-bit sampling at 20GSPS ... so why not just do the bandpass in the digital domain. The answers suggest the ADC must be protected from high signal levels. 

The below article from ADI describes the AD9084 functionality called "FFT Sniffer". This functionality (built-in to the AD9084) seems to suggest using a 512-point FFT to inspect the complete spectrum (0-18GHz) to detect a clean spectrum region where communication can occur. How does the chip do this w/o the hardware bandpass filter to protect the ADC? Does the LNA with AGC provide the necessary protection?

FFT sniffer discussion:   https://www.analog.com/en/resources/media-center/v...

see the discussion of IPhone FBAR filters (around 17:40 into the youtube article)

[ - ]
Reply by mfuerterJune 14, 2024

Jim, I understand what you're design goals are.  You are interested in development of a multi-band digital receiver.

The trade-off with such an approach is that interference from portions of the input bandwidth that are not being actively utilized will cause desensitization of the receiver.  Here I am assuming that the AGC [which protects the ADC from overload] has sufficient dynamic range to deal with the signal level deltas experienced in the use cases designed for. 

This is an example of an interference scenario that would reduce the receiver sensitivity: 

Let's assume that we have a 5G NR FR1 receiver [FR1 is the lower portion of the 5G allocations which extends up to 6 GHz].  Further let's assume that the UE [handset/phone] is receiving signal in the n41 band [2.5 GHz] from a serving cell that is 800m away at a signal level of -75 dBm/20MHz.  The effective input thermal noise floor in 20 MHz, assuming 6 dB receiver noise figure, is -95dBm/20MHz.  Thus, we have 20 dB SNR when no interference is present.

Next assume that the UE is still 800m from the serving n41 cell [-75 dBm/20MHz] and very close to a competitor's cell [which cannot serve the UE] which is transmitting n71 [~700 MHz] and the received signal level from the n71 cell is -50 dBm/5MHz [a level is easily found in every n71 network].  Now we have a problem, since the AGC will need to reduce the gain in front of the ADC to protect the ADC from overload.  This reduction in gain will increase the receiver noise figure and hence increase the effective input noise floor. Let's say that the AGC gain reduction results in an increase of the overall receiver noise figure to 12 dB [6 dB increase], now instead of the SNR being 20 dB it is only 14 dB. This has a big impact on performance since the the MCS [Modulation Coding Scheme] reduces at about 1 bit/Modulation_Symbol per 3 dB reduction in SNR.  If we reduce the spectral efficiency from 5.5 bits/symbol to 3.5 bits/symbol, then we have reduced the throughput by 36%...  

You can see that the wider the bandwidth allowed into the AGC/ADC, the higher the likelihood of such problems occurring. This problem is alleviated by utilization of RF filter sub-banding as you alluded to above.

Finally, I would likely use at least a FR [Frequency Range] selective RF filter to limit the input bandwidth to that which the device is use-cased.  For example for a FR1 UE, I would likely use a FR1 low pass RF filter.  But, maybe for low cost device you can get away without..

I'm writing a book here, so I'll stop.  Hope that this is of some use.              

[ - ]
Reply by jekain314June 15, 2024

Thanks Matt (and Chuck) for your thoughtful response(s). I am getting in to the communications field after "retiring" from another engineering field -- solely because of what I see with the chips like the AD9084 and other similar DirectRF chips from TI and NS. 

To your FR1 receiver example. There are three signals: the data at FR1=1.0, the receiver noise=1/100.0 (20dB SNR), and the competing large signal at N71 > 100.0. If the AGC introduces a gain of 0.01 to protect the ADC, then the FR1 signal becomes 0.01 and the receiver noise becomes 0.0001. So isnt the real issue the ADC dynamic range (e.g., 12 bits is 0-4095 so only 41 counts to handle the data). Doesnt the observed receiver noise get attenuated by the AGC? If we could do the ADC at 16 bits(0-65535) we would have 655 counts to work with.

But I see the problem with the large out-of-band neighbors. I see that there are tunable bandpass filters. This part, ADMV8526, will do 1.25-3.0GHz with 3dB rejection 9% of centerfrequency -- controllable with 8bit word. This would allow a dual-purpose 802.11 and GPS receiver. Tunable filter specs are below.




[ - ]
Reply by mfuerterJune 17, 2024

You're welcome.  Regarding your second paragraph:

The cascaded noise figure for a system is given by F_Cascade = F1 + (F2-1)/G1.  In this application the F1 can be considered the noise figure of the RF Preselector [RF Filter], the LNA, and the AGC. And F2 will be defined as the noise figure of the ADC.

I looked into the AD9084 [BTW an ADC that we could have only dreamed about 20 years ago] and the noise figure is about 32 dB [-150 dBFS, where FS is -2 dBm: dBFS = dB relative to Full Scale].  This is a big noise figure.

Now demonstrate how the SNR is degraded when the AGC is forced to respond to protect the ADC from large incident signal:

initially let's assume that all of the RF signal processing hardware in front of the ADC has a [cascaded] noise figure of 2.3 dB and that the gain of this "front end" hardware is 40 dB when there is no interference present.  Note that the RF frontend includes the AGC function.  Now using the cascaded noise figure formula we find that the aggregate system noise figure is 2.69 dB.  Here is a simplified illustration to help visualize:

Next, let's assume that we have a strong interferer present and the AGC must reduce the overall frontend gain by 20 dB. In this case we still have a frontend noise figure of about 2.7 dB, but the frontend gain is now only 20 dB.  The result of this gain reduction is to increase the impact/prominence of the huge ADC noise figure.  Now the aggregate system noise figure is 12.4 dB [an increase of almost 10 dB].  Here is another illustration to help visualize:

This 10 dB increase in noise figure will reduce SNR by 10 dB.  Note that this reduction in SNR will certainly be fully observed in thermal noise limiting environments and at least partially observed in Interference + Thermal Noise limited environments.  

The consequence of reduction in SNR [SINR] is that a lower order modulation scheme must be implemented to maintain the target BLER [Block Error Rate].  In general [5G or WIFI] the modulation-coding scheme curve follow closely with the optimal coding law and thus for every 3 dB of SNR [SINR] reduction the gNB will apply a lower efficiency modulation-coding scheme that is 1 bit/modulation_symbol less efficient.

I have attached a tool for you to use for these calculations.  You can play with this using different inputs.

Cascaded Noise Figure Calculator

If you have further questions, maybe we can have a call. This will likely be more efficient in terms of information transfer.

Forgive typos...


[ - ]
Reply by mfuerterJune 17, 2024

I noticed that the illustrations were dropped from my response above. The attachment below contains the entire response, inclusive of the illustrations.

ADC Cascaded Noise Figure Response + Illustrations

[ - ]
Reply by jekain314June 18, 2024

cant seem top open the attachments ..

when I click, nothing happens.

maybe just send to:  jekain314@waldoair.com

[ - ]
Reply by mfuerterJune 18, 2024

I just sent these via email...

[ - ]
Reply by ChuckMcMJune 14, 2024


Matt is correct here. You "can" skip the selector (the hard bandpass) but in doing so you get a higher noise floor and spectrum folding effects from the sampling. You really want a hard bandpass of the bandwidth you're sampling (slightly less if you have adjacent spectrum you are trying to avoid) to keep things that you aren't intending to look at in your data. So if you're sampling at 200MSPS centered around 2.45GHz you will will want a bandpass that blocks below 2.35GHz and above 2.55GHz. At that point you could use the built in FIR to isolate just the signal of interest.

Good rule of thumb here is that the more RF you can reject *before* you get to the ADC the better your performance will be with regard to resolving signals.

[ - ]
Reply by Mannai_MuraliJune 14, 2024

This is not my area.But I am asking to get more exposure.

Assuming we do not use RF filter interference close to 18 GHz can only alias and

affect the intended signal around 2.4 GHz.Others close to 10 GHz even if alias will be rejected by FIR.So Can I assume interference close to 18GHz will be very week to be considered.

[ - ]
Reply by WaveslingerJune 14, 2024

It depends on your application, but in the vicinity of airports, ports and so on, your receiver might be exposed to high level signals from Radars which would likely overload the converters so physical rather than digital front end selectivity is still a requirement if you want to maximize your dynamic range under all circumstances.

[ - ]
Reply by jekain314June 18, 2024
I am thinking of an app that does both comm and GNSS in the presence of unwanted (and unknown) signals. The AD9084 4Tx4Rx transceiver chip has a "Spectrum Sniffer/Monitor" so you can real-time see whats happening between 0-18GHz. I see parts called "RF Pre-selector" as well as digitally tuned bandpass filters. I havent found a part that does both the LNA and the RF bandpass to prep the signal for the ADC. 


[ - ]
Reply by WaveslingerJune 18, 2024

Take a look at https://www.atlantamicro.com/.

This company has a wide range of components specifically intended for applications such as these; digitally tuned filters, switched filter banks, amps, tuners etc, etc

You need to be aware though that tunable filters generally have a high insertion loss which will directly affect the system noise figure if placed in the best location which is right at the front end.

An alternative here is to have a high linearity, low noise amp at the front end followed by your tunable filter. This does make extreme demands on linearity of the LNA though and such amps are usually quite power hungry but gives a better noise figure overall.