Kali FIFO Buffer & Reclocker for SBCs

kali_on_bench

Back in August an interesting thread was started on diyaudio.com.

It described a FIFO buffer and reclocker, aimed at SBCs and more specifically the RPi.

Its name was Kali.

The FIFO board would be able to “fix up” the RPi’s problematic I2S output so as to improve the sound quality of the used DAC.

As the discussion progressed, more interesting details came to light.

Kali was basically an FPGA design with on-board RAM, high quality clocks and flip-flops. It would buffer the DATA stream in RAM (about 0.7 seconds of audio) and it would then reclock it using flip-flops outside of the FPGA. The clocks used for the reclocking would be high quality NDK units sporting extremely low phase noise.

It would be powered by a 5V/3A power supply and would supply filtered power to the RPi (or not, selectable by a jumper) as well to the DAC that would sit on top of Kali.

It would have a claimed 3ps of jitter, which is an impressive feat for any I2S source.

It would provide a high quality MCLK output from a U.FL jack underneath the board.

The board’s general availability was scheduled for the end of August, and its price would be in the neighbourhood of $70.

At around mid August cdsgames offered to give away a number of units to diyaudio members for testing. I took him up on his offer and he was kind enough to send me one (along with a Piano 2.1 DAC board, but that’s for another post).

Fast forward to mid-September, when I received a package from India. In it was Kali and Piano 2.1 (more about that in another post).

kali_in_packaging

On the left side of the board we can see the DC IN jacks (barrel and pin header), plus a couple of inductors that help with filtering the power lines. On the top left there is a jumper that controls whether Kali will supply power to the RPi or not. The Kali lists as minimum requirement a 5V/3A power supply. However, Kali itself will consume only about 100mA. The rest of the power is intended for the RPi and the DAC board. Note that Kali is designed so that it powers the RPi and not the other way around.

In the middle of the board the two chips that dominate are the 4Mbit SRAM chip and the Lattice MachXO3L-4300 FPGA itself.

To the right we have the two NDK clocks, one for each one of the two “families” of sampling rates.

kali_clock

The Kali comes in two versions. One with clocks at 22/24MHz, capable of supporting sampling rates up to 192KHz, and one with 44/48MHz clocks, going up to 384KHz.
Note that there are two footprints available for the clocks so that a curious (and somewhat experienced) DIYer can also try out clocks with different footprints, like for example Crystek units. If you choose to go that way, keep in mind that you will also need to change 2 capacitors from 0.01uF to 0.1uF.

At the top of the board we have the usual 2×20 header that reproduces the RPi’s GPIO pins. There is one notable exception – Kali does not supply 3.3V at the relevant pin, so if your DAC needs to take 3.3V from that pin it will not work (more on that later).

Note that even though Kali has a JTAG header, the manufacturer currently does not support firmware upgrades. That’s understandable.. JTAG is not for consumer use and the manufacturer has to protect his IP..

At the lower end of the board we have an array of LEDs indicating the sampling rate of the incoming signal, along with the status of the buffer (Empty / Full / Lock) and the selected oscillator.

kali_lights_on_44-1k

At the bottom of the board there is a U.FL socket that outputs Kali’s MCLK.

kali_mclk_out

So, we have an RPi, we have Kali, now what we need is a DAC. Kali will put a few restrictions on your DAC choices. Your DAC will have to run as “slave” to the RPi, that is it will have to take its BCLK from the RPi and not the other way around. Examples of such DACs include for example virtually all of the DACs based on ES9023 chips. DACs that are designed to run as masters (in order to battle the RPi’s legendary jitter problem), such as the Hifiberry Dac+ Pro & Digi+ will not work, at least not without some extra work.

Plus, like I mentioned earlier, Kali does not supply 3.3V on its top 40-pin header, so if your DAC requires that voltage you’ll need to find a way to supply it externally.

Straying a bit from the DAC HATs, Kali will have no problem feeding pretty much any DAC that accepts standard I2S, like for example a TPA Buffalo DAC.

kali_buffalo_iii

As a matter of fact, in my system Kali did a remarkable job of improving my RPi’s I2S output. The difference was obvious immediately. My RPi went from being a “net streamer quality” source to giving my Amanero a run for its money. Remarkable.
I am not talking about differences in bass or treble or tonality. It is like Kali manages to extract more music content from the data files. It revealed details that were there but were not audible. The music became more “real”. Imagine going from SD TV to HDTV. And that with the same 44.1K/16bit source material.

This improvement is not apparent only on expensive gear either. My first tests were done on my workbench with my RPi feeding an $8 9023-based DAC and listening through my headphones.

kali_tests_workbench

Even with this setup, Kali made a significant improvement to the sound. The $8 DAC went from sounding like an $8 DAC to performing at least decently. It was an obvious improvement.

So I can not imagine a scenario in which the Kali will not make an audible improvement to the sound.

Note that the first batch of Kalis (the ones that were sent to reviewers and a small number of units that were actually sold) had a bug that caused 16bit audio to have its channels reversed (and some more weird stuff happening with their I2S output resulting in somewhat degraded sound quality), but according to allo.com all effected units have been exchanged with fixed ones plus the ones that are shipping now to customers come with a fixed version of the firmware.

Even if for some reason you come by an affected unit, all you have to do is tell your RPi to output 32bit audio. That will fix everything.

I’d like to thank Ioan (cdsgames at diyaudio.com) for sending me the remarkable Kali. It will for sure become a permanent part of my audio chain.

At the time of this writing, Kali is being sold directly from its manufacturer’s site (allo.com) as well as from Volumio’s online shop.

The Raspberry Pi: Audio out through I2S

There are currently four ways to get audio out of the RPi:

  1. Use the audio out 3.5mm jack. It’s very easy to get it to work, but the sound quality is pretty bad, since it uses PWM to generate the sound. Due to that, its real resolution is in the neighbourhood of 11 bits. We have no use for that.
  2. Use the HDMI port. It works OK, but is useless to us audiophiles.
  3. Use a USB to I2S adapter, such as an Amanero or an XMOS-based device. Now we’re talking. They work quite well, and the quality of the I2S signal is dependent largely on the technology used (CPLD vs. XMOS, etc) as well as the quality of the on-board clocks. The problem is that they add another link to the audio chain, as well as increase the cost. Remember, the RPi is supposed to be a low cost solution.
  4. Use the GPIO pins of the RPi to get direct I2S output. This sounds way more interesting, right? Let’s try that!

According to several sources on the Net, this is the pin out:

Raspberry_Pi_B_Plus_I2S_out

You will probably notice that the RPi does not support MCLK output. This means in practice that your DAC will need to have its own on-board clock (or internal PLL / oscillator or whatever). We can live with that.

Luckily, my Buffalo III has its own clock (of course it does!) and thus can be connected quite easily. Let’s try that:

IMG_8297_resize

Now we have to configure the software for I2S output. For my distribution of choice, Archphile, it’s a piece of cake: http://archphile.org/howto/i2s-dacs-and-the-raspberry-pi/

Audio playback works just fine!

Well, almost fine..

You see, in theory the RPi has a bit of a problem with its I2S output. Since the only clock onboard the RPi is a 19.2MHz crystal, it should have trouble generating proper clocks for its I2S output. For example, for 44.1KHz audio, the LR Clock must be running at precisely 44.1KHz. That is not possible, since the frequency is not a multiple of 19.2MHz. Thus, the frequency can be either 19.200.000 / 435 = 44.138KHz or 19.200.000 / 436 = 44.0366KHz. This is a limitation of the Broadcom BCM2835 in conjunction with the 19.2MHz crystal and there is nothing that can be done.

In order to confirm the theory, I decided to run a few tests. I hooked up my logic analyzer to my RPi, set it up for I2S output, and fed it some 44.1KHz music.

IMG_8453_crop_resize

I took 1 sec worth of samples with my logic analyzer, configuring it for I2S signal. I got this:

logic analyzer 4

The PCM Clock is already appearing a little dodgy. Let’s zoom in:

logic analyzer 5

logic analyzer 6

As you can see, the pulses do not have the same duration. They appear to alternate between two values. So it is obvious that the signal has jitter. A lot of jitter. Since we’re here, let’s have a look at the LR Clock signal as well:

logic analyzer 7

logic analyzer 8

The duration of the pulses appears to alternate between 11.33μS and 11.38μS, giving respectively 44.12KHz and 44.04KHz, values very close to the ones I calculated previously.

So, the theory is sound and the RPi’s clock is not up to snuff by strict standards. What this means is that the RPi’s I2S output is not capable of “Hi End” audio transmission. It is essentially not bit perfect (edit: this is not correct, strictly speaking. It is in fact bit perfect, it is just not “proper”.).

In the real world, chances are that this problematic clocking will not be particularly audible under normal circumstances, say with a normal-specc’ed sound system. But an audiophile should definitely steer clear of the RPi’s I2S output, instead opting for a USB to I2S interface.