KCBS DRM finally decoded via AI, but partially. Facts:
– the MW+SW DRM network carries the “Third Program” (not 2nd!)
– They use a 13,2 kbps EVS codec in the DRM data channel – this carries the audio content
– The EVS codec is modified, but not encrypted. Need the modified decoder also.
– No regular MSC (audio channel)
So there are 2 audio files here.
1. stream recording from Koryo Front “KCBS-2” stream
2. partially decoded DRM signal from 3205 kHz. (MW signals had too many CRC errors when tried to decode)
The “Podo mida” (News) can be heard on both recordings at 00:24!
DRM datas:
Service Identifier: A00002
Service Name: 조선중앙제2라지오방송 (Korean Central Radio 2)
Country Code: kp, North Korea
DRM Mode and Bandwidth: Mode B, 10 kHz
MSC and SDC Modulation: 16-QAM / 16-QAM
Indicated Services: 0 audio services, 1 data service
Data stream bit rate (including internal encapsulation): 14.56 kbit/s
DEEPER ANALYSIS with help of AI:
After removing the header and checksum from the 716-byte data unit, 712 bytes of application data remain. A consistent structure is evident within this:
First field: 660 bytes – Exhibits a regular internal pattern; the hypothesized audio frames can be derived from this.
Next field: 2 bytes – Contains `00 00` in every unit examined.
Last field: 50 bytes – Variable content; partly related to a counter and a repeating cycle.
By prepending the first five bytes of the 660-byte field to the five subsequent 131-byte blocks, five 132-byte blocks are formed. Dividing these into four parts each yields twenty candidate frames, each 33 bytes long.
This aligns with one of the standard bitrates for EVS:
33 bytes × 8 bits / 20 ms = 13.2 kbit/s.
Indeed, a frame in the 13.2 kbit/s EVS mode consists of 264 bits, i.e., 33 bytes. (Based on etsi.org)
Beyond the frame length, the mode indicators also match. For each of the 5640 candidate frames obtained via the rearrangement described above, the first five bits take on a value corresponding to the superwideband (SWB) mode in the EVS 13.2 kbit/s mode table:
Value of the first five bits – Occurrence – Interpretation according to EVS
31 – 5585 – SWB, low-bitrate MDCT mode
10 – 33 – SWB, generic mode
12 – 15 – SWB, transition mode
11 – 7 – SWB, voiced mode
In the DRM IQ recording, at the positions where the test tones should be, each of the three short signals is accompanied by nine consecutive frames (starting with the byte `FE`) on either side. Interestingly, the entire 120.4–128.8 second temporal context of the signal consists of 21 consecutive data units with valid (error-free) CRC. There are no missing radio data blocks in this segment.
Nevertheless, the three 550 Hz tones and the single 1101 Hz pure tone cannot be reconstructed from this data. This represents an important step forward in troubleshooting: the issue with audio generation in this section cannot be explained by the loss of RF data; the interpretation of the application data and the assumed codec remain to be investigated. The original DRM was also attached!
Tringer László HNG (2026-11-15)