Carver TX-11b

The Carver TX-11b is a classic 1980s tuner that features FM stereo noise and multipath reduction, wideband AM, and AM stereo reception.

Including protrusions, the tuner occupies 19″ × 14″ × 3⅞″. You can remove the handles for rack mounting, but the hole spacing is nonstandard. The tuner weighs 11Ľ lbs and consumes 15 W.

The display is vacuum fluorescent and all controls are pushbuttons. 13 memories store frequency and mode.

Circuit

An F-connector provides 75Ω antenna input. The RF amplifier and mixer use dual-gate MOSFETs. A single tuned circuit before the RF amp and three that follow yield a flat RF passband when sweep-aligned. Although the front-end uses AGC, it easily overloads. The wide IF uses a pair of Murata SFE10.7MX-A 250-kHz ceramic filters and two adjustable IF transformers. Narrow cascades three SFE10.7MZ1-A 180s and another transformer with the wides. All filters are the blue, low-group-delay type. An LA1235 IF amplifier / quadrature detector feeds an LA3401 stereo decoder, which uses a ceramic resonator and no oscillator trimpot. Unbuffered LC ultrasonic filters drive the audio output, which high-voltage 4966 CMOS switches route. Series/shunt JFETs provide muting. Back-to-back Zeners protect the output circuitry.

AM tunes 520–1710 kHz. A 5″ × 2″ loop antenna has an inductance of 13 µH and a Q of 74 at 1000 kHz. It drives a fix-tuned RF transformer. A switchable RF attenuator made from a resistive optoisolator feeds a JFET/bipolar cascode RF amplifier. A double-tuned circuit precedes the mixer. The 450-kHz IF has two bandwidths. An LA1135 integrates several RF and IF functions, while a Motorola MC13020 provides C-QUAM AM stereo reception. The frequency response extends to 15 kHz in wide. (The tuner was designed before FCC rules limited audio to 10 kHz.) Response is 3 dB down at 5.7 kHz in narrow. An audio equalizer flattens the response. A front-panel switch invokes NRSC AM deemphasis. A 10-kHz notch filter kills adjacent-channel heterodynes, while a 25-Hz stereo pilot canceller enables deep bass response. A patent covers several of the AM circuits, including a very effective noise limiter.

The controller/PLL is a Toshiba TC9147BP. A switch on the PLL board selects 100-kHz FM steps and 10-kHz AM steps for the US, or 50/9 kHz for Europe. A switch on the main PCB selects 50- or 75-µs deemphasis. A jumper in the power supply permits 110 or 220 V operation. The supply uses three-terminal regulators.

A TX-11a block diagram is here. The TX-11b antenna connections and front-end differ from those shown.

Noise and Multipath Reduction

The TX-11b uses a patented analog technique to reduce noise and multipath distortion in FM stereo. It replaces the L−R signal with a quieter and cleaner three-component signal. The first component is delayed and spectrally shaped L+R whose level is proportional to that of L−R. The second component is flat L−R whose level is squared. The third component is 480-Hz lowpass-filtered L−R with its level reduced about 8 dB. Noise reduction uses roughly equal contributions from the first two components, while the first component dominates for multipath reduction.

L+R is about 23 dB quieter than L−R so the first component provides a low-noise source of ambience with no directional cues. Panasonic MN3007 bucket-brigade devices delay L+R 25.6 ms below 3000 Hz and 10 ms above. Although their S/N spec is 80 dB A-weighted, the TX-11b surrounds each BBD chip with complex volume companding circuitry. LM3080 operational transconductance amplifiers provide the variable gain. The 10-ms path uses preemphasis/deemphasis. Nine-pole lowpass filters in each path suppress the BBD clock. To make its level proportional to L−R, the circuit multiplies the delayed L+R signal by the ratio of the L−R and L+R levels. Op-amps and diodes perform the division using logarithms. A diode bridge attenuator performs the multiplication.

The second component is L−R whose level is multiplied by that of single-pole, 400-Hz, highpass-filtered L−R. At high levels this component provides strong directional cues and masks any noise. As L−R drops, this component drops even faster to suppress background noise that might become exposed with little to mask it. The level squaring essentially behaves as a soft squelch. An LM3080 performs the multiplication. The high-pass filtering reduces this component's contribution without altering its spectrum when the signal consists mostly of lower frequencies.

According to the patent, the third component “contains little of the undesired noise and is much less susceptible to multipath distortion.”

The circuitry greatly reduces noise and multipath distortion while still providing a stereo effect. The improvement is impressive, often recovering an unlistenable signal. But the sound can be somewhat distant, diffuse, or reverberant, especially for multipath reduction. Acoustic images may be displaced, particularly with headphones. Still, I prefer the stereo presentation to mono or noisy stereo for most program material. Although not nearly as transparent as the DSP noise reduction in the Sony XDR-F1HD, the Carver circuit was a remarkable achievement when introduced in the early 1980s.

Pressing MULTIPATH REDUCTION increased the third harmonic of a 1-kHz stereo test tone to 1.1%. The distortion was easily audible on the sine wave, but I've never noticed any on program material. Pressing NOISE REDUCTION instead increased the third harmonic to 0.11%. The distortion originates in the diode bridge attenuator.

Carver called the noise reduction circuit an Asymmetrical Charge Coupled Detector. The delay circuits use MOS charge-coupled devices. The circuit is a processor, not a detector, and I could find nothing asymmetrical about it.

Understanding the block diagram and schematics was easier once I realized that many of the annotations were misleading or incorrect. The BBD delay values are wrong. COMPARATOR is a divider. AGC AMPLIFIER is a voltage-controlled attenuator. A.G.C'ed LINE AMP does not use AGC. 450Hz PASSBAND FILTER is a 480-Hz lowpass filter. 480Hz HI-PASS FILTER is −3 dB at 600 Hz including the parallel path. 3KHz LO-PASS FILTER is −3 dB at 2.8 kHz and 3KHz HI-PASS FILTER is −3 dB at 3.9 kHz (the two filters do provide equal contributions at 3.0 kHz). LEADING EDGE DETECTOR neither favors leading edges nor performs detection. RF GAIN +12dB removes attenuation rather than adding gain, and the change can be much greater than 12 dB.

I measured this response for the delayed L+R signal from the noise reduction input to just prior to the output stage. I sampled the interference dips due to summing the two delay paths only from 2.0 to 3.6 kHz where they were most pronounced. 0 dB is arbitrary.

The patent says that a spectral shaping stage with this response makes the delayed L+R spectrum more closely match that of L−R. The curve does not include everything in the delayed L+R signal path that affects the spectrum, such as the 600-Hz highpass filter that greatly alters its overall shape.

This is the response of the SPECTRAL SHAPING AMP from a circuit simulation. Compared with the curve in the patent, the depression is centered at twice the frequency and is 2.7 dB deeper, and the high-frequency shelf is 5 dB lower.

The TX-11b noise reduction differs in other ways from that described in the patent. The Leading Edge Detector in the patent really does favor leading edges. R232 in the L−R gain-control path is 160kΩ. In conjunction with C85 and R233, a 16-ms highpass time constant boosts leading edges by 11 dB as shown in the blue circuit simulation curve above. In the TX-11b, R232 is 6.2kΩ, which provides the negligible 0.8-dB rise shown in the red curve. The Leading Edge Detector is the second noise reduction component described above. Finally, the patent contains complex logic circuitry that bypasses the noise reduction depending on composite signal quality. The TX-11b does not include this circuitry.

This is the frequency response of the noise reduction output stage from a circuit simulation. It affects unprocessed L+R as well as noise-reduced L−R. This may be the reason the owner's manual says that the noise reduction can “improve” L+R 1 dB. I found no mention of this final spectral adjustment in the patent.

This is the noise reduction circuitry. The SIP ICs are dual op-amps. A schematic is here.

AM Noise Limiter

The TX-11b uses a simple but highly effective AM noise limiter. The diodes clip positive and negative excursions beyond the signal envelope to suppress impulse noise. The charging time constant is long enough that the capacitor voltages follow the signal envelope, not noise spikes. The 100kΩ resistors slowly discharge the capacitors so that the clipping thresholds automatically track the audio level. The diodes do clip waveform peaks, but the effect is minimal. Forgetting to disable the noise limiter has little audible consequence.

Pin 3 comes unglued from ground when the op-amp output saturates during a clip. This adds R357 to the input voltage divider, lowering the input attenuation and causing a slope discontinuity in the nonlimited waveform passed to IC47 pin 8. But the effect is small and may not be audible. To eliminate it altogether, short R349 (and R350 for the channel not shown) and realign the AM audio output level.

The AM deemphasis circuit follows the limiter and attenuates clipping harmonics.

Modifications

To improve stereo S/N and to eliminate HD Radio self-noise, I installed a postdetection filter on the underside of the main PCB. The filter reduces noise without incurring the artifacts of the Carver noise reduction system.

I added a .0013-µF phase compensation capacitor from LA3401 pin 3 to ground to improve stereo separation. Holes and pads exist for such a part, but neither the schematic nor silkscreen show it. To bring the separation trimpots within range with the postdetection filter installed, I added 3kΩ across R113 and 3.9kΩ across R114.

I could not quite tune the 7.2-MHz PLL reference oscillator on frequency. TC501 wound up set all the way to its lowest capacitance. I tried unsoldering one lead of C318, 2.2 pF across TC501, but it didn't help much. I lifted one lead of C319, 47 pF on the other side of the crystal, and added 47 pF in series. That worked. The crystal is marked and specified as 7.199 kHz, but generating correct LO frequencies requires oscillation at 7.200 MHz. Perhaps 7.199 kHz is the series-resonant frequency and a parallel-resonant crystal was mistakenly supplied.

Adjacent-channel selectivity for the original 180-kHz narrow filters was 24 dB. While this was a great improvement over the wide filters, it still wasn't enough for the crowded FM band in my area. I replaced the filters with two 110s and a 150. I selected the individual filters for symmetrical modulation-induced noise to minimize sensitivity degradation. Adjacent-channel selectivity improved to 52 dB, which is the highest figure I've ever measured for an analog IF strip.

The TX-11b provides no way to force monophonic reception. Mono reduces noise without adding artifacts. It's particularly useful with stations that retain the stereo pilot during monophonic programs. You can add a resistor from the AM deemphasis switch to the stereo decoder to force mono when the pushbutton is out. Normally you'll leave the button in for FM stereo and AM deemphasis. The modification stops the 38-kHz VCO by applying +12 V through 39kΩ to LA3401 pin 17. A half-watt resistor has leads long enough to install directly between the switch PCB and the main PCB. Solder it to the front left deemphasis switch terminal as shown.

I discovered that a strong local AM signal caused spurs in the FM band. Grounding the insulated RCA audio output jacks to the rear panel eliminated the interference.

The original AM circuit used a fixed 120-pf capacitor across the antenna transformer secondary. Resonance was much too narrow to cover the entire band and off-resonance loss directly degraded the receiver noise figure. To improve sensitivity, I removed the capacitor and added a varicap diode between the tuning voltage at C002 and the transformer secondary. Tracking was not that good for the junkbox varicap I used, but sensitivity still improved over much of the band. I replaced C002 with a high-quality ceramic with lower impedance in the AM band to eliminate some spurs coupled by the varicap. With this modification I can receive signals on any AM channel at night with the stock loop.

Pressing NOISE REDUCTION or MULTIPATH REDUCTION blanks the audio for about 1.4 seconds. I found this delay annoying during rapid A/B tests (see Sound Samples). I tried reducing the blanking interval by changing the RC value on the switch PCB, but occasionally a loud pop occurred. I reluctantly restored the original blanking interval.

The unbuffered audio output impedance is rather high and rises with frequency. To minimize high-frequency roll-off, avoid high-capacitance output cable and equipment with excessive RFI input filtering. To eliminate the problem altogether, add op-amp output buffers. You can use a relay or JFETs to increase the buffer gain in narrow IF mode to compensate for the lower output level (3 dB lower for the narrow filters I used).

Alignment

TX-11a AM and FM alignment instructions are here, with changes for the TX-11b here.

Align the main PCB with the noise reduction board unmounted and folded over to the left. Run a ground lead to the unmounted board to align AM, which passes signals through it. The board is grounded only through one of its standoffs.

Instead of alignment steps 7 and 8, I swept the FM front-end and greatly improved the RF passband shape. Input return loss was remarkably constant at 9 dB across the entire FM band.

The alignment instructions do not mention T601. This tuned circuit on the front-end board in parallel with the IF signal path can help minimize stereo THD (I used it mainly to suppress the third harmonic). The mixer transformer and two IF transformers on the main PCB also affect stereo THD. These adjustments provide so many degrees of freedom that I was able to make the distortion products disappear in wide IF mode. (Later they rose somewhat when the temperature changed. I report that figure below.)

Noise reduction alignment instructions are here. Though not explicitly stated, they assume that the signal generator uses preemphasis. To align without it, alter the audio levels in steps 8, 13, 27, 30, and 32. Use the deemphasis calculator in this set of tuner measurement utilities. Step 34 specifies an absolute RF signal level. Check for the expected noise behavior somewhat above and below the specified level since the front-end alignment and noise figure affect where it occurs.

A 1987 service bulletin applies when “customer complains of too much echo with the ACCD circuit engaged”: align to 75 mV in step 19. This lowers the delayed L+R component 9.5 dB.

A 1989 TX-11a application bulletin changed the AM deemphasis to the new NRSC standard: R403 and R404 should be 3.3kΩ and C161 and C162 should be .0033 µF. These components are in the upper left corner of the noise reduction PCB. This change was supposed to have been incorporated in the TX-11b, but mine (S/N A90700999) had the old values. The circuit simulation curves above compare the two responses. The blue curve is the new deemphasis and the red curve is the old.

When I first aligned the 10-kHz AM notch filters, I neglected to preset RP23 and RP24 as instructed. I simply adjusted all trimpots for the deepest notch at 10 kHz. I got the response shown above for the two channels. The frequency span is 5–15 kHz at 1 dB/div.

After presetting the trimpots and realigning the filter, I got this response. Notch depth was > 45 dB.

Instead of peaking the IF transformers in steps 2 and 3, I adjusted them for the most symmetrical response while sweeping the IF. I noticed that the IF passband was somewhat rounded, which made me wonder how the audio response could be so flat. I hadn't noticed the stereo equalizers in the AM audio circuit. The circuit simulation curve above shows their response with the single adjustment at midsetting. Instead of step 15, I swept the modulation and set the adjustment for the flattest tuner frequency response from 500 Hz to where the notch filter takes over above 9 kHz. Then to compensate for any small channel difference in deemphasis, I enabled it and slightly readjusted one channel so that the response curves coincided over most of the frequency range (at 1 dB/div they were indistinguishable to 8 kHz).

Measurements

For the following measurements I used IEEE 185-1975, updated as described here. I used the test equipment listed here. The tuner had all of the modifications described above except output buffers and shorted noise limiter resistors. I did not implement the 1987 service bulletin.

50-dB quieting sensitivity, mono     W 14.4 dBf, N 15.2 dBf
50-dB quieting sensitivity, stereo   W 36.0 dBf, N 36.0 dBf
THD, 1 kHz, stereo                   W < 0.03%, N 0.3%
THD, 1 kHz, stereo NR                W 0.11%
THD, 1 kHz, stereo MR                W 1.1%
Stereo separation, 1 kHz             W 48 dB, N 44 dB
S/N, 65 dBf, mono                    83 dB
S/N, 65 dBf, stereo                  75 dB
S/N, 85 dBf, stereo                  77.5 dB
Capture ratio                        W 0.7 dB, N -0.4 dB
Capture ratio, stereo                W 13.5 dB, N 14.0 dB
Capture ratio, stereo NR             W 7.5 dB, N 7.0 dB
Adjacent-channel selectivity         N 52 dB
RF intermod                          66 dBf (roughly 20 dB worse than for other tuners)
RF spur                              99 dBf
RF image                             91 dBf
RF AGC threshold                     72 dBf
RF mismatch loss                     0.6 dB
Noise figure, 96.9 MHz               3.9 dB
Modulation acceptance                W 280%, N 160%
Minimum stereo pilot injection       2%
Treble response                      W +0.1/-2.2 dB (-0.3 dB to 14 kHz)
Bass response, -1 dB                 < 10 Hz
Output level                         W 0.57 V, N 0.40 V
Output impedance, 15 kHz             5kΩ
AM THD, 1 kHz, 95% AM, deemphasis    W < 1%, N < 3.5%
AM deemphasis error to 9.0 kHz       L +0.8/-0.5 dB, R +0.5/-0.5 dB
AM deemphasis error to 9.5 kHz       L +0.8/-0.6 dB, R +0.5/-1.0 dB

Model Differences

The TX-11 and TX-11a differ greatly, while the TX-11a and TX-11b are nearly the same.

The TX-11 receives FM only and provides both 75Ω and 300Ω antenna inputs. The front-end has four tuned circuits between the RF amplifier and mixer, an IF transformer cascaded with the mixer transformer, and no RF AGC. The LA1235 feeds an LA3381 stereo decoder with oscillator trimpot through an LC postdetection filter. There are 16 station memories. Audio muting uses series JFETs only and the power supply uses discrete regulators. Some noise reduction component values differ significantly from those in later models and the TX-11 includes the logic circuitry of the noise reduction patent.

The TX-11a uses the TX-11 front-end. It is otherwise the same as the TX-11b, except for the AM deemphasis time constant for some units and appearance parts.

Sound Samples

This sound sample illustrates how multipath reduction affects a signal with severe co-channel interference. The blanking interval is due to the tuner. This sample demonstrates wideband AM reception.


August 20, 201588–108 MHz