Improving the noise floor of a Delta44 A/D converter
(Dec 1 2002)

The ground loop error

The Delta 44 soundcard is sensitive to loops in the ground connections between the computer and the analog radio hardware. Even with careful star grounding it is difficult to avoid strong signals at mains frequencies. The computer itself may produce pretty large signals due to potential differences between different parts of the chassis. The ground loop is easily eliminated by cutting the connection between the two screws holding the 15 pin d-sub connector and the analog ground plane of the PCB itself. Look here for performance measurements for different modifications of the Delta 44.

Power supply noise

With the high dynamic range of the A/D converters it is utterly important to have extremely well decoupled supply voltages. performance measurements for different modifications of the Delta 44 shows the modifications I found suitable on my own Delta 44 card when adding them one by one. Modifications interact and the final solution is a little simpler but the result is the same (see below).

The input amplifier/filter

The Delta 44 uses JRC 5532 op-amps for both input and output. For the input, the 5532 is configured as a simple lowpass filter that attenuates by 20dB at 1.2MHz while it attenuates by 0.8dB at 48kHz. Due to the limited size of the coupling capacitors the input amplifier also serves as a high pass filter with the 3dB point at about 3Hz.

The input amplifier adds about 1dB to the noise floor. When using the Delta 44 together with reasonably well designed radio hardware the input amplifier is superfluous. The AK4524 codec used in the Delta 44 has built-in filters, a high pass filter at 0.9Hz (DC error cancellation) and a RC filter at 64 times the Nyquist frequency. The codec samples at 64 times the specified sampling speed and uses digital filters to produce a steep anti-alias filter at the Nyquist frequency. The 5532 low pass filter is probably intended to give additional suppression at about 3MHz where the oversampled AK4524 has false responses although they are probably not very strong due to the built-in RC filter.

By replacing the input amplifier/filter by two resistors and an electrolytic capacitor the noise floor is lowered by 1dB and the phase shift at low frequencies is reduced which makes calibration easier at very low audio frequencies. (The center discontinuity can be removed with less accurate data i.e. with fewer pulses collected in the calibration procedure)

Modifications interact

In a study to evaluate the feasibility of supplying modified Delta 44 soundcards to the amateur radio community I have evaluated several boards and checked how the modifications interact. As it turns out, the decoupling of the 7805 is less critical when the input amplifiers are bypassed. Using 470 microfarads for C25, C61, C45, C47, C51 and C53 is more than adequate.

The digital power supply

The 10 microfarad capacitors from pins 23 (digital power supplies) of the AK4524 codec chips have not only the desired effect but these capacitors also couple the noise from the digital ground plane to the analog ground plane through only 2.5 ohms. One solution is to change R15 and R16 to 15 ohms from the value of 5 ohms recommended by the manufacturer. Another solution is to remove R15 and R16 completely and feed the codecs by wires from the unmounted IC U7. Pin 14 of U7, connected to C38, gets 5V directly from the PCI bus. A 470 microfarad in the C37 position makes this 5V supply well decoupled. The data sheet states that the analog supply should be powered up before the digital supply and that the digital supply must not exceed the analog supply by more than 0.3V. This means that this solution may be risky in case the 12V computer supply fails while the 5V supply is still running. Removing the analog supply voltage while the digital supply is kept at 5V will destroy the codec immediately!

A 1N4007 diode from pin 14 of U7 to the output pin of U5 ensures that the analog supply is never more than about 0.6V below the PCI 5V supply. 15 ohm resistors in the digital supply lines to the codecs set the digital supply voltage 0.3V below the PCI supply. Hopefully these two precautions will protect the codecs of the modified board against power supply failures. I have NOT tested all possible error conditions and there is no guarantee that the modified board will survive all possible power supply failures.

Noise floor comparisons

In total six boards have been tested. Four of them have been modified. As can be seen from table 1 the boards differ somewhat in their original state but after modification they are practically identical. The noise floor is measured with nothing connected to the input and it is measured in a bandwidth of 1kHz. The numbers in table 1 are recalculated to 1Hz bandwidth and expressed as dB below the strongest non-saturating sinewave that the codec can handle.


Board     Original          Modified
         (-dBc/Hz)         (-dBc/Hz)  
 no      L        R         L      R
  1     144.3   144.3     148.4   148.5
  2     146.0   145.9     148.5   148.5
  3     145.4   145.4     148.4   148.5
  4     146.3   146.0     148.4   148.4
  5     145.9   146.0
  6     146.3   145.9
Table 1. Noise floor at 30kHz of modified and unmodified Delta 44 boards when no signal is present expressed as dB below the strongest possible carrier in 1 Hz bandwidth. Note that 30kHz is the signal frequency in I/Q mode. Two audio channels are combined to give the spectrum of one RF signal.

As can be seen in table 1 the modifications reduce the noise floor by 2 to 4 dB. A quick comparison between fig. 1 and fig. 2 shows that the modifications improve performance significantly more than indicated by table 1 when a strong signal is fed to the input. The signal source used to produce figures 1 and 2 was the 2.5MHz to audio converter that is a part of the high performance hardware for linrad.

Here is a step by step description of the Delta modifications and a high resolution image of the modified board



Fig. 1 Unmodified board (no 6) with a near saturating signal at 2.508MHz using the 2.5 MHz to audio converter as the signal source.


Fig. 2Modified board (no 3) with a near saturating signal at 2.508MHz using the 2.5 MHz to audio converter as the signal source.