Selectivity improvement solutions proposal for UmTRX

Jean-Samuel Najnudel - BJT PARTNERS SARL jsn at bjtpartners.com
Sun Mar 18 14:25:33 UTC 2012


Hi Andrey,

Thank you very much for these answers. Please see my comments bellow.

On Sat, Mar 17, 2012 at 11:31 PM, Andrey Sviyazov <andreysviyaz at gmail.com>wrote:

> Hi Jean-Samuel!
>
> Some answers and reasons below.
>
> Regarding the choice between HMC483MS8GE/HMC485MS8GE and
>> HMC686LP4E/HMC687LP4E, I cannot figure out which one would be the most
>> suitable for our design. For the filter, it is also quite difficult for me
>> to help you to decide between the TB0130A and the TB0448A.
>>
> By the way, do you think this would be fine to use a different mixer for
>> 900 and 1800 MHz bands ? As chip packages remain the same, I think it would
>> be fine but I would like to double check with you.
>>
>  Differences between HMC 483/483 and HMC686/687:
> 1/ price of HMC483/485 twice lower.
> 2/ current consumption of HMC483/485 twice lower.
> 3/ package ms8g better for me.
> 4/ I have good expirience with HMC483/485 and some qty there are on my
> warehouse in Moscow.
> 5/ HMC483/485 require TB0130A which have 20dB higher rejection than
> TB448A.
> So, HMC483/485 better for me.
>

This looks fine for me. However, I cannot see 20 dB higher rejection for
TB0130A, may be 5 or 10 dB, not more. Anyway, this is still better.
Just to double check, could you please confirm low ripple bandwidth of the
TB0130A is wide enough for a GSM carrier ?
If yes, your choice looks perfect for me.



>
> Regarding the GSM spec, I believe these blocker tests are hard to pass and
>> not that useful in most practical situations. However, I do not care that
>> much about passing this spec for my network deployment in Mayotte but I
>> really believe passing the spec will be very important for you if you wish
>> to sell your hardware solution to some major operators. Moreover, as we
>> would anyway need a superheterodyne selective filtering to get a reasonably
>> narrow Rx sampling band (< 1.5 MHz LMS band), it does not cost that much
>> more to try to pass the GSM spec.
>>
> You right, it is important for us and may be it is really important for
> systems with high channel dencity.
> Furthermore, it is very easy to lose the reputation of the product, but it
> is very difficult then to fix it back.
> I think, low cost doesn't sign low quality, so we must to have good
> hardware on market for good sales and promote OpenHW :)
>

I agree. :-)



>
>
>> By the way, I really believe this does not worth designing 2 different
>> preselectors. I think it would be more reasonable to design only the
>> Hittite based high performance preselector. First of all, as you mentioned
>> and as I calculate on my side also, the ADRF6601 will not even pass the
>> micro BTS spec. Moreover, the time you spend designing and debugging 2
>> different preselector would probably cost more than the 40 USD per board
>> you save on the boards you will use for low cost applications. Moreover, 2
>> different designs would mean lower quantities components order. This would
>> mean higher buying prices, higher shipping costs per chip... In practice,
>> the cost difference would be probably 25 to 30 USD per board, not 40 USD.
>>
>
>> I really think we should only make one design with the Hittite high
>> performance preselector. If we really want a low cost femtocell, a
>> universal SDR board or a lab experiment system, we would not even populate
>> the preselector components. If we need GSM deployment applications (except
>> femtocells), nobody (even me ;-)) would actually care about the extra 40
>> USD BOM.
>>
> Exactly, I agree again.
>

:-)



>
>
>> I looked at the picture of your Rx path calculations. As discussed
>> together in Barcelona, I am not sure we should add an LNA on the UmTRX. TMB
>> LNA would already have a very good Noise Figure (about 0.8 dB or even a bit
>> less) and a quite high gain (between 20 to 30 dB depending of the cable
>> loss). OIP3 of the TMB would be typacally around 35 dBm. Adding an extra on
>> board LNA would be quite bad for the total Rx path IIP3.
>>
> How you define that 35 dBm OIP3 required?
> I saw only -1dBm IIP3 in the datasheet for middle Gain of LNA2 and it
> isn't informative for me.
>

35 dBm OIP3 is not really required. I would say, the highest is the best as
long as price is reasonable. High OIP3 would avoid IMD interferences
problems.
I already have some offers, products and prototypes from Chinese
manufacturers. The OIP3 of the LNA they can provide is around 35 dBm.



>
>
>> Moreover, as there is some PCB leakage from Tx to Rx in the UmTRX, we
>> might avoid onboard amplifiers. An on board amplifier might actuallly
>> amplify the leakage.
>> I actually suggested this LNA a few months ago but, in Barcelona, you
>> convinced me this LNA was not really a good idea. What do you think ?
>>
> Duplexer+PA(0.25..2W)+LNA+Preselector are all will be on the mezzanine
> board for uBTS, nBTS and mBTS solutions TMD.
> So, you right - new UmTRX will be without LNA's and bandpass filters,
> because it is universal dual tranceiver with diversity switch.
> As we discussed in Barcelona, we forced to design different mezza-boards
> for each band, but it much easy than UmTRX changes.
> Also high power "mezza-boards" will be not as true mezzanine because of
> require heatsink and separate power source.
>

I agree for band specific filters (to avoid too many UmTRX versions). I
agree for LNA (to avoid amplifying the Tx leakage inside the UmTRX PCB and
to avoid a too high gain that can lower our IIP3). I agree for the PA (for
all the reasons you mention).

However, I really believe the preselector should remain on the UmTRX main
PCB.
If we want to connect the preselector as an external mezzanine board, it
would cost us 5 extra SMA connectors (RX1in, RX1out, RX2in, RX2out, clock
source), 2 RF switches before the superhet filtering (to still be able to
have switched diversity), several GPIO (serial mixer control, RF switches
control) and power supply.
This is possible but it would really cost much more than keeping the
preselector on the main PCB, especially as I believe, in practice, we will
need almsot always this preselector, even not to pass the GSM spec but just
to be able to get a reasonably narrow sampling bandwidth.
I would really suggest to keep the preselector in the main PCB and
populating or not the components (VCO/PLL, mixer and SAW IF).



>
> Regarding the PA parameters, for both models we can get at a good price,
>> P1dB would be between 8 (Infineon) to 15 (Freescale) Watts at 27 VDC (a bit
>> lower at 24 VDC). Gain would be around 40 dB (excluding cable loss). Please
>> let me know if you need others LNA and/or PA specs of the TMB I can get
>> from the manufacturers I am discussing with.
>>
> Actualy I really thought that you already have offers about PA or TMB from
> Chinese manufactures.
>

Yes, I have some interesting offers. I am still discussing with them but
they propose nice products for affordable prices.



>
> Best regards,
> Andrey Sviyazov.
>
>
Best regards.

Jean-Samuel.
:-)


>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osmocom.org/pipermail/umtrx/attachments/20120318/dde1b6ba/attachment.html>


More information about the UmTRX mailing list