[opendtv] DTT tuner design (was Re: Demand for free DTV rising in Australia)

I haven't been reading the 'Australia' thread until just now, but I see it's become a continuation of the old ATSC receivability debate in disguise. Most interesting to see the very thorough FCC tuner test report though.
But still some of the old misconceptions are bouncing around in here, about single vs double conversion receivers, tracking filters and the rest. Please can we lay them to rest once and for all?
1. "Double (or dual-) conversion inevitably confers a phase noise penalty" -- Wrong!
Yes, of course phase noise is additive, but that doesn't mean you end up with more! It's all a matter of taking advantage of the conversion scheme. Back in 1997 I designed a DVB-T tuner for BBC R&D, to demonstrate the results of their collaboration with LSI Logic on COFDM demodulators. The phase noise spec was tough, <-90dBc/Hz at 10kHz offset, a performance unattainable at that time with a single VCO/PLL covering the UHF band (almost an octave) at the required 167kHz (precision offset) steps.
We created a double-conversion design using Anadigics (GaAs) up- and downconverter devices with Zarlink (then Mitel) synths. First IF sat at 1100 MHz with a pair of 3-pole ceramic filters for image rejection, with second IF at 36MHz using European cable-TV type SAW filters. We achieved image rejection in the 70dB region, and adjacent-channel protection ratios of about -40dB. Channel matrix was 8 MHz, so we used a 4MHz increment for the upconversion and 167kHz for the downconversion. The upconverter's wide tuning range (and thus high open-loop phase noise) VCO was tamed by the large step size (low division ratio) and wide loop bandwidth to achieve some -95dBc/Hz at 10kHz offset. The downconverter for its part only needed to tune a handful of MHz (to account for VCO circuit spreads) so could use a more conventional narrow loop and also achive -95dBc/Hz. Result: -92dBc/Hz combined.
In fact that tuner had a third downconverter: the LSI device needed to sample at the low IF of 4.57MHz, so we had a crystal-controlled mix-down from 36MHz; this didn't materially affect overall phase noise. 

2. "Double conversion tuners don't need a tracking filter" -- Wrong!
This BBC design was never intended for consumer or field use, only for demonstration and for lab and transmitter measurements. For this reason we dispensed with front-end filtering, but implemented an autonomous wide-band RF AGC to control average signal level at the first mixer. Together with the use of GaAs devices here, this ensured the required high IP3 (region of +12dBm). The tuner was designed into the BBC's own professional monitoring receiver, licensed to Broadcast Technology for manufacture as the DVTM-2000(T). It also formed part of the BBC's own radio camera system.
Quite predictably, when connected to an antenna in a high interference environment, its front-end gain is reduced by the wide-band AGC. This blocking effectively reduces sensitivity in the presence of any out-of-channel (but in-band) interferer above the RF AGC threshold (about -45dBm). For this reason it was not suitable for consumer or field applications, but unfortunately it was the only product which could be modified for the 6MHz channel matrix in time for the flawed Sinclair tests of 1999. For later demonstrations in Korea we applied a front-end tracking filter to excellent effect.
When I was approached in 2001 to design a consumer DVB-T tuner, I chose a similar architecture, but this time implemented an alignment-free tracking filter ahead of the front-end AGC. The GaAs mixers were swapped for integrated up- and downconverter chips in silicon (Zarlink, now Intel) with a (single) SAW-filtered first IF at 1220MHz and a single SAW at 36MHz, the final IF. In addition to the RF AGC (now filtered and with a limited range) we implemented a delayed AGC at the high first IF, as well as the main (2nd IF) AGC, giving us fine control over gain distribution to maximize dynamic range. The tuner card had a BoM cost of about $12, but this included the COFDM demodulator (NxtWave, now ATI) and PAL remodulator (Motorola, now Freescale).
This became the SetPal-1 product, marketed under the NovaPal, Dijam, Daewoo, Labgear, Triax and British Telecom brands, which set new standards for DTT STB RF performance and became for a while the reference against which the DTG test facility (at BBC R&D) judged other products.
3. (a) "Single conversion is necessarily inferior in interference rejection" -- Wrong! 
3. (b) "Image-reject downconversion isn't used in consumer products" -- Wrong!
In 2003 I redesigned the SetPal tuner for single conversion, as a cost-reduction measure. Out went the double-conversion scheme, and instead I used a new Zarlink part, the SL2610. This was a single-conversion TV tuner chip with an image-reject mixer. Phase noise wasn't nearly as good as in the double-conversion design, but still left us a good 6dB margin. The degree of image rejection offered (at the n+9 taboo channel, equivalent to America's n+14 -- remember we use 8MHz channels with a 36MHz IF) wasn't sufficient to deliver the performance I needed, so I augmented it with an alignment-free RF tracking filter. In order to escape the requirement for production alignment this was necessarily wider than the 12MHz or so normally used in analog tuners: 3dB bandwidth was some 50MHz, with a notch some 40dB deep over 10MHz around (signal + 2xIF). The 36MHz IF used a single SAW filter with an IF/AGC amplifier. Volume BoM cost about $6.50, including again the COFDM demod (Zarlink MT352) and UHF analog re-mod as before.
SetPal-2 outperformed the double-conversion SetPal-1 on every test parameter (e.g. ACPR -45dB, 64QAM sensitivity -83dBm), despite it's being less than half the cost. It replaced SetPal-1 in the STBs mentioned, and was also used successfully in DVRs, iDTVs and iDVCRs by Beko, DigiFusion, Goodmans, Matsui, Bush and Daewoo. 

4. "Cheap DTT tuners don't employ a tracking filter" -- Wrong!
Inexpensive as SetPal-2 was, the UK Freeview market became increasingly cutthroat after 2003, and the $150 price point for STBs soon fell to a $50 expectation. Daewoo's Northern Ireland operation became unprofitable as supply shifted to Korea and China, where (in addition to huge economies in main board design) manufacturers used the cheapest tuners they could get, with little regard for performance. These were can types, weighing in at a dollar and a half, consisting of a standard (Infineon or Philips) TV tuner chip, using the vendor's reference design, with a tuned MOSFET AGC RF stage and tracking filter ahead of it. There was no IF section in the can: instead the 36MHz IF came out onto the host PCB where a Chinese SAW filter and inexpensive IF amplifier were inserted, before the COFDM chip's ADC.
The tracking filter was the same as analogue tuners had used for years: two poles ahead of the MOSFET, two behind, varactor-tuned. Usually realized with one- or two-turn coils, or little loops of wire, manually bent in production alignment until the sweep fit the template: a matter of seconds per unit -- cheap in China.
But the result was good enough: the low DTT transmitter power and high analog interferer levels (ahead of DSO) have conditioned the British public into expecting to need a new antenna installation for 'digital'. So a whole generation of RF-inferior products flooded the Freeview market. And maybe it worked OK in the UK, where rooftop antennas generally look at co-sited transmitters and most consumers 'know' they shouldn't expect to 'get digital' on a set-top antenna. But try and use the same tuners in the US and there will be tears.
Which is at the root of this debate. And it's such a shame that the US-market products skimp on RF performance -- America never had a SetPal -- as the price point is arguably less important there than in Britain: what's an extra 4 or 5 dollars on the cost of a TV, if it means the off-air function works reliably? Instead the demod chip vendors are jumping through generations of hoops trying to score an extra dB of decoding margin, while the tuners are throwing away those same dBs in tens.
Intel still offers the descendents of SL2610. They also have some excellent demod devices for MRC diversity (in COFDM, though perhaps not yet in 8-VSB -- I haven't checked). The new silicon tuner devices (Microtune, Maxim, Broadcom, NXP, Freescale, Xceive etc.) typically include combinations of RF tracking filter, double conversion and image reject mixers, either for real-IF or zero-IF architectures, and they're now beginning to cope with realistic off-antenna signal and interference levels too. So the potential is there for a recovery in typical RF performance without cost-up, if the vendors have the heart to employ it.
If you want to build a receiver that works both in the boonies and the concrete canyons, get the tuner right!
[As a solution to the Freeview set-top antenna limitation for 'second set' applications, I currently have here in the lab a 4-way diversity system: 3 electronically tuned active antennas in an LCD screen surround, plus an external antenna jack. Four Freescale tuner chips (double-conv with discrete filters, I'll probably go for more integration next time and eliminate those filters), Intel demods, single board. Typically 8 to 10dB indoor diversity gain. To say any more would look like a commercial!] 

Stephen J Birkill



----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org 
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word 
unsubscribe in the subject line.

Other related posts: