Who makes the best highly linear RF amplfier that meets this spec?

I'll probably get this wrong ;)

2V/20bit ~= 2uV/bit

2uV/sqrt(100e6) ~= 0.2nV

-Lasse

Unfortutunately, those that have grown up in the digital era thinks that adding 1 or 2 bits will just add very small 10-20 % incremental cost in signal processing paths.

The real (analog) world is not that forgiving, Any additional bit in ADC/DAC will require an additional 6 dB in SNR and in the worst quadrupling the power consumption.

Thus, you really gave to understand what to do in the analog domain and know what makes sense to do in the analog domain (_very_ expensive) and what can be done in the digital domain (cheap).

Unless you understand these tradeoffs, you unfortunately are going to design suboptimal (non-competive) products.

yes, and your statment is true, but how to apply to what I'm doing??

If you put "RF amplifier" in the heading, no wonder that everyone thinks about radio receivers or transmitters :-).

I still do not understand all your requirements.

You wanted 100 MHz bandwidth but use only 200 Msamples/s, but did you think to implement the anti-alias filter ? There are no such thing as brick wall filters. You either have to limit the bandwidth to say 80 MHz or increase the sampling rate to 250 Msamples/s.

Your input amplifier output requirement +14 dBm is about 1 Vrms into

From other posts, it appears that you are only interested of a less than 1 MHz bandwidth somewhere below 100 MHz. Assuming you want to run the ADC at 100 MHz bandwidth and digitally filter out a 1 MHz segment, you really do not need a 18-20 bit ADC, since the input noise will dither the signal quite nicely. A 14-16 bit ADC should be sufficient, but of course the digital filtering must be done at 24 to 32 bits.

Tricky.

-- john, KE5FX

Not with low distortion at 100MHz.

Only if there is wacks of loop gain at Fo times N. To get -90db distortion your talking multi 10s GHz effective FUG. i.e to reduce the second harmonic of 100 MHz, you need wacks of gain at 200 MHz!

A 3Ghz ft device will possibly get you a loop gain of 3, if your lucky.

shielded from each other.

I agree that achieving low distortion at 100Mhz+ probably requires ASICs.

Kevin Aylward B.Sc.

Kevin Aylward B.Sc.

Or: Chris Trask noiseless feedback

each.

the

makes

amps

other

The TI IC THS6182 seems to be a dsl line driver and may be fast enough. Perhaps check out similar devices.

?-)

wow, thanks, I allocated 10nV/rtHz, and 80 dB distortion so looks like a doable part!

I hat those URLS, my system goes to the website and sits there for days never getting anyfurther to get the data sheet!

a google search gave a specific .pdf url, which sadly doesn't yield a .pdf directly, but slowly gets one. eventually. so got it.

Oh lordy, i would like to hear this story. I once was a test engineer.

?-)

doing susceptibility testinghere that would be verified in England WITHOUT being able to attend, so had to design the tests to be 'more than' the requirement. To meet 30V/m sometimes hit the unit with 100 to 200 V/m using a huge rack mounted AR and an antenna bombarding and an antemnna behind monitoring.

The owner of the lab was performing the test for me and put he amplifier into standby for me to go fiddle with the telco unit. When I stepped inside the room the flexible ground plate on the chamber's floor flexed sending the AR into self oscillation at some unknown frequency at unknown power level. Before he could shut down, it blew out a ceiling flourescent fixture about 15 ft aboce inside hte chamber, and continued to arc down the leg of the tripod holding the monitoring antenna [long enough to burn a distinctively large hole through the carpet from the leg to the flexible ground plate, AND evaporated the 20dB, 20W HP attenuator in series to the spectrum analyzer. We took it apart and found nothing inside. SA? can't rembmer, but I think good old HP front end survived.

The AR manual clearly mentioned that standby only works if you don't violate the Vswr, gave a number, which obviously happend with the combination of room, floo, and antenna.

I could have been killed. but didn't even notice except for the sizzling sound and the popping of the luminarie and cascading down shower of sparks.

By the way, the telco company I worked with was the ONLY bidder on the contract to demonstrate to BABT in England, at their test site, to meet ALL the electrical specs of performance. We beat two major japanese firms two major US firms, two major European suppliers, and the preferred UK company. And, didn't get the contract! went to [guess who?] the preferred UK company. Based upon their 'promise' to meet all specs. And then the UK company showed up at my client's doors six months later, 'offering' to buy their technology! That's a lot of gall!

People have had problems with TI's scripts before:

Tim

TI's website used to be incredible. used to go straight to the part, all the appnotes, models, etc. now....just get a page 'promising' all that, but never 'completing' the page [always an hourglass on it], worse never getting passed that page, can't even select anything on it.

For the really 'problematic' websites I use the other systems, which are left dormant, ie, turn on the WinXP with IE, or more useful Firefox, to get though to those websites. But dislike those systems entirely. Although these are high powered PC's operating higher powered OS; they run like molasses compared. Plus, the 'out of date' combination of Opera, on Win98, through dial-up has proven some of the best virus/malware protection. An attack either stalls the PC, or better yet crashes it to a stop. And if something starts tying to 'diddle' the PC, just reach down and unplug the telco line before enouch code can get in/out. Happened only three times, but handy. Plus, makes for the built-in 'interference' of that bloated code required to be constantly running on the system to be far less intrusive.

so far, people have kindly provided 'straight to' URLs that allow me to download in the background [without bogging down the PC] while continuing using the PC. All the new websites bog, or halt, the PC rendering a 'web surfing' to be a single process, taking often an 30 mintues to 45 mintues to do what a 5 minute telco does. like get name and address of closest store. Oddly, anymore when I call and ask to get the phone number AND address closest to me, the delay seems overly long. They often render apologies of, "my computer is slow today" yest, everybody seems to be saying that anymore. Say what? aren't they on DSL, or high speed and using those high powered PC's? yet, they seem almost as slow as my access.

thanks again for the 'heads up' on using that TI part. 3.2nV/rtHz, compared to allocated 10nV/rtHz, is like increasing the dynamic range by 3 times! The overall effect improves by only 20%, but 20% is 20% and no other way to get there.

Err..., unfortunately not. The distortion quote at ***1 MHZ*** was the immediate clue.

At a gain of 5, its -60db @ 10MHz. It doesn't plot at all for > 15 Mhz

The amp is no way near fast enough for 100 MHZ operation, nor is it low distortion at that unobtainable frequency.

i.e. Its S.R. of 450/us @ Av=5, means 70 Mhz max at 1V output, at which point its distortion is of the chart.

I pointed out ADC drivers, because that they are specifically designed to be fast with very low distortion, i.e. things wanting N bits of accuracy.

DSL line drivers are not designed for that class of performance.

Kevin Aylward B.Sc.

I stand corrected. I 'glanced' at the data sheet. That's my excuse and I'm sticking to it.

yes, ADC drivers have the best potential to work well. Don't need much gain, so should be possible.

enough.

which

to be

Damn, now i am going to actually read the datasheet. What you are saying it can do is at odds with i understand that it must do. DSL often sends a lot more data than just a few Mb/s of network, it often carries transport streams for many TV channels all at the same time. Sending 100 Mb/s or more of data on a 9 mile line takes some doing.

On page 4 the bandwidths looked a lot better than what you said but the output swing in rather small and the slew rate may be only 45 V/uS, check the graphs though.

The really bad news is on page 8, it pretty well dies at after about 30 MHz.

Oh well.

?-)

h.

o be

-Lasse

See my other post.

I am up on this because I have spent some time looking at the design of ADC unity gain S/H buffers. Getting 16 bits at 100 MHz is very difficult. Making an amplifier with gain, is harder. The 2.8 GHz small signal BW of the LMH6554 implies, possibly, something like 50 GHZ Fts for its transistors.

As you go up in frequency it becomes proportionately harder still, 3rd order will go up at least at a square order rate.

In fact considering H = BW (log 1 + S/N), for digital multilevel data transmission, where N is actually error not just noise, including the issue of distortion by assuming F=BW, error =k.F^M

H = F. log(1 + S/(k.F^M))

It should be clear that increasing F (BW) will result in less data rate than at lower frequencies. The error is increasing at a rate faster than the normal root(f) associated with noise based errors.

Kevin Aylward B.Sc.